<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Elif TUTAR on Medium]]></title>
        <description><![CDATA[Stories by Elif TUTAR on Medium]]></description>
        <link>https://medium.com/@eliftutar007?source=rss-77dd05013bb8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*qa5JAVIzNkqqLVnVFR59BA.jpeg</url>
            <title>Stories by Elif TUTAR on Medium</title>
            <link>https://medium.com/@eliftutar007?source=rss-77dd05013bb8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 18:38:45 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@eliftutar007/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[5.317 Komut, Tek Operatör: Yapay Zekâ Siber Saldırıyı Nasıl Yeniden Tanımlıyor?]]></title>
            <link>https://medium.com/@eliftutar007/5-317-komut-tek-operat%C3%B6r-yapay-zek%C3%A2-siber-sald%C4%B1r%C4%B1y%C4%B1-nas%C4%B1l-yeniden-tan%C4%B1ml%C4%B1yor-962633ffa4a8?source=rss-77dd05013bb8------2</link>
            <guid isPermaLink="false">https://medium.com/p/962633ffa4a8</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[yapay-zeka]]></category>
            <category><![CDATA[ai-security]]></category>
            <dc:creator><![CDATA[Elif TUTAR]]></dc:creator>
            <pubDate>Sun, 13 Sep 2026 12:16:40 GMT</pubDate>
            <atom:updated>2026-09-13T12:44:04.899Z</atom:updated>
            <content:encoded><![CDATA[<p>Check Point’ten Anthropic’in yayımlanan yeni tehdit istihbaratı raporu: <strong>2026&#39;da saldırı, savunma ve güven krizi aynı anda büyüyor.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8RYt76vEveoS4puvmrWJuA.png" /></figure><p>Geçenlerde farklı alanlarda çalışan arkadaşlarımla yapay zekânın işimizi nasıl değiştirdiğini konuşuyorduk. Konu siber güvenliğe geldiğinde soru bir anda tersine döndü: <strong>Biz AI ile saatler süren işleri dakikalara indiriyorsak, saldırgan neden indirmesin?</strong></p><p>O an mesele benim için daha netti. Yapay zekâ sadece saldırıları daha gelişmiş hâle getirmiyor; <strong>saldırının zamanını, maliyetini ve gereken insan emeğini de azaltıyor.</strong></p><h4><strong>Peki ne kadar?</strong></h4><h3>Saldırı Tarafında Maliyet Çöküyor, Uzmanlık Bariyeri Kayboluyor</h3><p>Saldırı tarafındaki en çarpıcı değişim maliyet çöküşünde görülüyor. Check Point’in raporuna göre bir geliştirici, yapay zekâ ortamı kullanarak 88.000 satırlık uçtan uca bir komuta-kontrol saldırı çerçevesini (VoidLink) <strong>bir haftadan kısa sürede</strong> üretti — geçmişte bir ekibin aylarca sürdürebileceği bir iş. Zamanlama tarafında da benzer bir sıkışma var: <strong>bir zafiyetin kamuya açıklanmasıyla ilk exploit’in ortaya çıkması arasındaki süre günlerden saatlere indi;</strong> bazı ülkelerde düzenleyiciler bu nedenle zorunlu yama süresini 12 saate kadar çekmek zorunda kaldı.</p><p>Ölçek de dikkat çekici: Check Point’in 2026 Cyber Security Report’una göre kuruluşlar 2025&#39;te haftada ortalama <strong>1.968 siber saldırıya</strong> maruz kaldı — 2023&#39;ten bu yana <strong>%70&#39;lik bir artış</strong>. Aynı raporda kurumsal veri sızıntısı riski de büyüyor: yüksek riskli GenAI istemlerinin oranı geçen yıl içinde %2&#39;den %4&#39;e ikiye katlandı, kuruluşlar ayda ortalama 10 farklı yapay zekâ uygulaması kullanıyor — çoğu resmi onay almadan. Anthropic’in kendi tehdit istihbaratı ekibi bu değişimi net bir cümleyle özetliyor: az kaynaklı, teknik derinliği sınırlı gruplar artık büyük, iyi finanse edilmiş ekiplerin işini yapabiliyor.</p><p><strong>Türkiye açısından bu tablo soyut değil.</strong> Kaspersky’nin META bölgesi için yayımladığı 2026 ilk yarı verilerine göre, tehdit aktörleri kötü amaçlı yazılım geliştirme sürecinde harcadıkları süreyi ve maliyeti büyük ölçüde düşürerek çok daha geniş ölçekte faaliyet yürütebiliyor — ve Türkiye, bölgede saldırıların en yoğun hedeflendiği ülkelerden biri. Bu, yapay zekâ siber güvenlik tehdidini demokratikleştirdi, sadece büyütmedi anlamına geliyor.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RewVWYTjUC7gZARtApjuGQ.png" /></figure><p><strong>Peki bu hız artışı sadece saldırganların mı işine yarıyor?</strong> Hayır — aynı ivme, savunma tarafında da benzeri görülmemiş bir ölçeklenmeye kapı açıyor.</p><h3>Siber Güvenlikte Yapay Zekâ Savunması: Zero-Day Avı ve AI SOC</h3><p>Project Glasswing, Anthropic’in kritik altyapıyı korumak için başlattığı savunma girişimi, saldırı tarafındaki bu hız artışının simetrik bir yanıtı olabileceğini gösteriyor. Başlangıçta yaklaşık 50 ortakla başlayan program, Haziran 2026 itibarıyla 15&#39;ten fazla ülkede <strong>150 kuruluşa</strong> genişledi; ortaklar şimdiye kadar <strong>10.000&#39;den fazla</strong> yüksek/kritik önemde zafiyet tespit etti. Ancak bu tablo eksiksiz değil: güvenlik camiasından bazı isimler, bulunan zafiyet sayısıyla fiilen yamalanan zafiyet sayısı arasındaki farkın hâlâ çok büyük olduğuna dikkat çekiyor — yani asıl darboğaz artık<strong> “bulmak” değil “doğrulamak ve yamalamak.”</strong></p><p><strong>XBOW </strong>örneği bu dönüşümün bireysel ölçekte nasıl göründüğünü gösteriyor: <strong>otonom bir yapay zekâ ajanı HackerOne liderlik tablosunda #1 sıraya yerleşti ve Microsoft Bing altyapısında CVSS 9,8 puanlı, kimlik doğrulama gerektirmeyen bir uzaktan kod yürütme (RCE) zafiyetini insan müdahalesi olmadan buldu.</strong></p><p>AI SOC (güvenlik operasyon merkezi) tarafında da yapay zekâ siber güvenlik yaklaşımının ne kadar ölçeklendiğini gösteren benzer bir tablo var — ama burada biraz daha temkinli olmakta fayda var. Gartner’ın Eylül 2026&#39;da güncellediği öngörüye göre, büyük SOC’ların %70&#39;i 2028&#39;e kadar Tier 1–2 operasyonlarında yapay zekâ ajanlarını pilot olarak deneyecek, ancak yapılandırılmış bir değerlendirme yapmadan bunların yalnızca %15&#39;i ölçülebilir bir iyileşme elde edecek. Gartner ayrıca 2030&#39;a kadar SOC ekiplerinin %75&#39;inin, otomasyona aşırı bağımlılık nedeniyle temel analiz becerilerinde erime yaşayabileceği konusunda uyarıyor — yani AI SOC gerçek bir dönüşüm ama sihirli bir değnek değil. Türkiye’de de bu dönüşüm somutlaşıyor: Türk Telekom, yapay zekâ destekli sistemlerle her ay binlerce saldırıyı otomatik olarak tespit edip engellediğini açıkladı.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Cxjbl8ilnf-X-3V3lsYjww.png" /></figure><p><strong>Ama bu simetri tam değil; çünkü savunmayı güçlendiren aynı ajan mimarisi, kendi başına yeni bir zafiyet kaynağı da yaratıyor.</strong></p><h3>Ajanların Kendisi Yeni Bir Saldırı Yüzeyi Oluşturuyor</h3><p>Araç kullanabilen, otonom hareket eden her yapay zekâ ajanı, aynı zamanda kandırılabilir bir hedef haline geliyor. <strong>OWASP’ın 2026 LLM Top 10 listesinde</strong> ilk sırada yer alan indirect prompt injection (dolaylı istem enjeksiyonu), tam olarak bu riski tanımlıyor: saldırgan modele doğrudan komut vermiyor, işlediği bir belgeye, e-postaya ya da web sayfasına gizli bir talimat yerleştiriyor; ajan bu içeriği “güvenilir” sanıp talimatı kendiliğinden yürütüyor.</p><p><strong>EchoLeak (CVE-2025–32711)</strong> bunun en somut örneği: Microsoft 365 Copilot’a gönderilen tek bir e-posta, kullanıcının hiçbir tıklama yapmasına gerek kalmadan hassas kurumsal verinin dışarı sızmasına yol açtı. Ölçek de küçümsenecek gibi değil: Google’ın verilerine göre kötü amaçlı dolaylı istem enjeksiyonu içeren web sayfalarının sayısı dört ayda %32 arttı. Check Point’in kendi tespitleri de bunu doğruluyor — uzun, kötü amaçlı istem enjeksiyonu payload’larının tespit oranı Mart-Mayıs 2026 arasında <strong>yaklaşık beş kat arttı</strong> ve gözlemlenen isteklerin neredeyse %1&#39;ine yaklaştı.</p><p><strong>Sonuç net:</strong> bir ajana e-posta okutmak, doküman özetletmek ya da web sayfası taratmak, artık sadece verimlilik değil, aynı zamanda saldırı yüzeyi anlamına geliyor. Araç kullanabilen bir yapay zekâ, aynı zamanda kandırılabilen bir yapay zekâdır — ve bu gerçek, yapay zekâ siber güvenlik stratejilerinin merkezine oturmalı.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T70AVB8NWaDXr3SvEPFB7Q.png" /></figure><p><strong>Peki</strong> bu ikili tehdit — hem hızlanan saldırılar hem yeni saldırı yüzeyleri — karşısında <strong>insanın rolü tam olarak nereye evriliyor? </strong>Ve şirketlerin birbirine bile güvenmediği bir ortamda bu denetim nasıl kurulacak?</p><h3>AI-vs-AI Çağında İnsanın Rolü Ne Olacak?</h3><p>Anthropic’in yayımladığı rapor, bu sorunun ne kadar karmaşıklaştığını gösteriyor. Rapor yalnızca dış saldırganları değil, yapay zekâ şirketlerinin birbirine karşı davranışını da gözler önüne seriyor: Çinli yapay zekâ şirketleri<strong> Moonshot AI ve DeepSeek’</strong>in, kullanıcı taleplerini kendi modelleri yerine gizlice Claude’a yönlendirdiği, kullanıcıların kendi modellerinin yanıtını aldıklarını sandığı ama aslında Claude’un ürettiği metni gördükleri tespit edildi. <strong>Alibaba bağlantılı aktörler ise 3.500&#39;den fazla sahte hesap üzerinden günde yaklaşık 3 milyon istekle, Anthropic’in “şimdiye kadar kaydedilen en büyük yasa dışı damıtma saldırısı”</strong> dediği bir operasyon yürüttü — rakip bir modelin yeteneklerini izinsiz kopyalamak için.</p><p>Aynı raporda, Yemen’in kuzeyinde faaliyet gösteren bir grubun <strong>güdümlü roket ve füze programlarında Claude Code’u navigasyon ve simülasyon yazılımı geliştirmek için kullandığı, </strong>Çin merkezli bir aktörün ise bir <strong>anti-torpido silah sisteminin ateş kontrol yazılımını geliştirmek amacıyla Claude’dan yararlandığı belirtiliyor </strong>— her iki hesap da tespit edilip kapatıldı. Bu vakalar, tehdidin artık yalnızca veri hırsızlığıyla sınırlı olmadığını, fiziksel dünyaya kadar uzanabildiğini gösteriyor.</p><p>Bu tabloda ortaya çıkan paradoks, yapay zekâ siber güvenlik alanının tam kalbinde yatıyor: aynı teknoloji hem saldırı maliyetini düşürüyor hem savunmayı ölçeklendiriyor hem de şirketler arası güveni zedeliyor. <strong>Anthropic’in GTG-1002 vakası</strong>nı çözerken izlediği yöntem, bu karmaşada insanın rolünün nereye evrildiğini gösteriyor: saldırıyı analiz eden ekip, büyük hacimli log verisini incelemek için bizzat Claude’u kullandı — insan ise stratejik yargıyı, doğrulamayı ve nihai kararı elinde tuttu. <strong>Yani rol “işi bizzat yapmak”tan “hedefi belirlemek, sınırı çizmek, stratejik yargıyı üstlenmek”e kayıyor.</strong></p><p>Milli İstihbarat Akademisi’nin (MİA) yakın zamanda yayımladığı<strong> “Yapay Zeka Çağında Siber Güvenlik ve Türkiye’nin Stratejik Öncelikleri”</strong> raporu da bu noktayı destekliyor: yapay zekâ sistemleri saldırı örüntülerini insana kıyasla çok daha hızlı tespit edebilse de, bağlamı yorumlama, yanlış pozitifleri ayıklama ve kritik kararları doğrulama kapasitesinin hâlâ insan uzmanlarda kaldığı vurgulanıyor.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OYBzPs7T5CwLNX9s2m5e4A.png" /></figure><h3><strong>Sonuç</strong></h3><blockquote><strong><em>Yapay zekânın siber güvenlikte asistan olmaktan operatöre dönüşmesi, ne saf bir tehdit ne de saf bir çözüm hikâyesi. Check Point’in verileri, Meksika vakası, Project Glasswing’in bulguları ve Anthropic’in dün yayımladığı rapor aynı gerçeği farklı açılardan gösteriyor: hız artık her iki tarafın da ortak dili, güven ise giderek daha kıt bir kaynak. Kazanan taraf, teknolojiyi elinde bulunduran değil, onu doğru sınırlarla, doğru yönetişimle ve doğru insan denetimiyle kullanan taraf olacak. Nitekim Milli İstihbarat Akademisi’nin de vurguladığı gibi, Türkiye özelinde bu denge, ulusal kapasite ve kurumsal karar kalitesiyle doğrudan bağlantılı. Bu da bizi başta sorduğumuz soruya geri götürüyor — mesele artık “yapay zekâyı kullanmalı mıyız?” değil, “onu hangi sınırlarla, hangi gözetimle kullanıyoruz?”</em></strong></blockquote><p><strong>Sizin kurumunuzda agentic AI güvenliği şu an nasıl ele alınıyor? Bu dönüşümü nasıl karşılıyorsunuz, yoksa henüz gündeminize hiç girmedi mi? Yorumlarda paylaşın, birlikte tartışalım. Siber güvenlik ve Yapay Zek</strong>â<strong> alanındaki gelişmeleri kaçırmamak için abone olmayı unutmayın.</strong></p><h3>Kaynaklar</h3><p>Bu yazıyı hazırlarken faydalandığım kaynaklar:</p><ol><li>Check Point Research, <em>AI Security Report 2026</em> (Temmuz 2026) — <a href="https://research.checkpoint.com/2026/ai-security-report-2026/">research.checkpoint.com</a></li><li>Check Point, <em>AI Security Threats in 2026: Annual Insights</em> — <a href="https://blog.checkpoint.com/ai-security/ai-security-threats-in-2026-insights-from-check-point-research/">blog.checkpoint.com</a></li><li>Check Point, <em>Cyber Security Report 2026</em> (Ocak 2026, haftalık saldırı istatistikleri) — <a href="https://www.barchart.com/story/news/37273720/check-point-softwares-2026-cyber-security-report-shows-global-attacks-reach-record-levels-as-ai-accelerates-the-threat-landscape">Barchart / GlobeNewswire</a></li><li>Anthropic, <em>Detecting and Countering Misuse of AI: September 2026</em> — <a href="https://www.anthropic.com/threat-intelligence-report-september-2026">anthropic.com</a></li><li>Anthropic, <em>Disrupting the First Reported AI-Orchestrated Cyber Espionage Campaign</em> (GTG-1002, Kasım 2025) — <a href="https://assets.anthropic.com/m/ec212e6566a0d47/original/Disrupting-the-first-reported-AI-orchestrated-cyber-espionage-campaign.pdf">assets.anthropic.com</a></li><li>Anthropic, <em>Project Glasswing: An Initial Update</em> — <a href="https://www.anthropic.com/research/glasswing-initial-update">anthropic.com/research</a></li><li>Anthropic, <em>Expanding Project Glasswing</em> — <a href="https://www.anthropic.com/news/expanding-project-glasswing">anthropic.com/news</a></li><li>XBOW, <em>XBOW Raises Series B to Advance Autonomous Penetration Testing</em> (HackerOne #1 detayı) — <a href="https://xbow.com/blog/series-b">xbow.com/blog</a></li><li>Aim Security, <em>EchoLeak — Zero-Click AI Vulnerability in Microsoft 365 Copilot</em> (CVE-2025–32711) — <a href="https://www.aim.security/lp/aim-labs-echoleak-m365">aim.security</a></li><li>Google Security Blog, <em>AI Threats in the Wild: The Current State of Prompt Injections on the Web</em> — <a href="https://blog.google/security/prompt-injections-web/">blog.google/security</a></li><li>Help Net Security, <em>Gartner: 70% of SOCs Will Pilot AI Agents. Only 15% Will See Results</em> (Eylül 2026) — <a href="https://www.helpnetsecurity.com/2026/09/09/prophet-security-evaluating-ai-soc-agents/">helpnetsecurity.com</a></li><li>Sabah, <em>Türk Telekom Siber Saldırılara Karşı Yapay Zekâ ile Kalkan Örüyor</em> — <a href="https://www.sabah.com.tr/yazarlar/sirt/2026/07/30/turk-telekom-siber-saldirilara-karsi-yapay-zeka-ile-kalkan-oruyor">sabah.com.tr</a></li><li>Anadolu Ajansı, <em>Kaspersky META Bölgesinin Siber Tehdit Görünümünü Açıkladı</em> — <a href="https://www.aa.com.tr/tr/isdunyasi/bilisim/kaspersky-meta-bolgesinin-siber-tehdit-gorunumunu-acikladi/703679">aa.com.tr</a></li><li>Anadolu Ajansı, <em>Milli İstihbarat Akademisi, Yapay Zekâ Çağında Siber Güvenlik Raporu Yayımladı</em> — <a href="https://www.aa.com.tr/tr/gundem/milli-istihbarat-akademisi-yapay-zeka-caginda-siber-guvenlik-raporu-yayimlandi/3916431">aa.com.tr</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=962633ffa4a8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Yapay Zekâ Karar Verirse]]></title>
            <link>https://medium.com/@eliftutar007/yapay-zek%C3%A2-karar-verirse-e85b97f82513?source=rss-77dd05013bb8------2</link>
            <guid isPermaLink="false">https://medium.com/p/e85b97f82513</guid>
            <category><![CDATA[chaos-engineering]]></category>
            <category><![CDATA[digital-twin]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Elif TUTAR]]></dc:creator>
            <pubDate>Tue, 13 Jan 2026 21:01:44 GMT</pubDate>
            <atom:updated>2026-01-14T06:48:20.061Z</atom:updated>
            <content:encoded><![CDATA[<p>Autonomous Agentic Ai, Digital Twin’ler ve Chaos Engineering</p><blockquote><strong>Bu yazı, yapay zekânın yalnızca öğrenmediği; karar vermeye başladığı noktada, sistemleri nasıl tasarlamamız gerektiğini ele alıyor.</strong></blockquote><blockquote><em>Yapay zekâ sistemleri bugün sadece doğru çıktılar üretmiyor.<br>Aldıkları kararların sonuçlarını izliyor, bu sonuçlara göre davranışlarını güncelliyor ve kendi sınırlarını test edebilecek yapılara dönüşüyor.Bu dönüşümle birlikte yazılım, sabit kuralların ötesine geçiyor. Karar alan, tepki veren ve kendini gözlemleyen bir sistem davranışı ortaya çıkıyor.</em></blockquote><blockquote><strong>Asıl soru artık şu: Karar verebilen sistemler, kendi davranışlarını nasıl test eder?</strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*ZG7biv57EFwOZtiAihROfQ.png" /></figure><h3>Karar Veren Sistemler Ne Anlama Geliyor?</h3><p>Bir sistemin karar vermesi, yalnızca “doğru cevabı seçmesi” anlamına gelmez.<br> Karar; bağlamı anlamayı, olasılıkları değerlendirmeyi ve sonucu sahiplenmeyi gerektirir.</p><p>Bugün<strong> autonomous</strong> <strong>agentic ai </strong>olarak tanımlanan yapay zekâ sistemleri tam olarak bunu yapıyor.</p><p>Bu sistemler:</p><p>Belirli bir hedef doğrultusunda hareket eder.</p><p>Ara adımlar planlar.</p><p>Sonuçları ölçer ve stratejisini belirler.</p><p>Bu noktada yazılım artık statik bir yapı değildir.<br><strong>Davranışı olan bir sistem</strong> haline gelir.</p><p>Ve burada önemli bir kırılma yaşanır: Karar veren bir sistem, aynı zamanda <strong>sorumluluk üreten</strong> bir sistem olmaya başlar.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*KwyMXk1x112SsXe6cTbeUA.png" /><figcaption>Autonomous Ai Architecture</figcaption></figure><h3>Kararlar Nerede Test Edilir? Digital Twin Yaklaşımı</h3><p>Karar verebilen bir sistemin en büyük riski, ilk hatasını gerçek dünyada yapmasıdır.</p><p>İşte bu noktada <strong>digital twin</strong> yaklaşımları devreye girer.</p><p><strong>Modern digital twin’ler:</strong></p><p>Sistemlerin birebir dijital temsillerini oluşturur.</p><p>Olası senaryoları simüle eder.</p><p>Kararların sonuçlarını önceden görmemizi sağlar.</p><p>Bu sayede sistemler:</p><p>Hatalarını üretim ortamında değil, <strong>simülasyon ortamında</strong> yaşar.</p><p><strong>Karar vermek kadar, kararın sonucunu önceden görmek de bir mühendislik becerisidir.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2nCUwWPfaP3nZuKRr-UvxQ.png" /><figcaption>Digital Twin Architecture</figcaption></figure><h3>Chaos Engineering: Sınırlar Bozulmadan Öğrenilmez</h3><p>Bir sistemin dayanıklı olduğunu varsaymak, onu dayanıklı yapmaz.<br>Dayanıklılık, ancak sınırlar zorlandığında sistemin verdiği tepkiyle ortaya çıkar.</p><p><strong>Chaos engineering</strong>, sistemleri rastgele bozmak değildir.<br>Amaç, kontrollü bozulmalarla sistemin gerçek davranışını gözlemlemektir.</p><p><strong>Agentic AI</strong> ve <strong>Digital Twin</strong>’lerle birlikte düşünüldüğünde:</p><p>Sistem kendi zayıf noktalarını keşfedebilir.</p><p>Dayanıklılık ölçülebilir hale gelir.</p><p>Öğrenme, hata sonrasında değil <strong>öncesinde</strong> gerçekleşir.</p><p>Sistem nerede kırıldığını öğrendiğinde, nasıl ayakta kalacağını da öğrenir.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*CyfNB7hj1WJp47f8oNWi0A.png" /><figcaption>Chaos Engineering Sample</figcaption></figure><h3>Karar Veren Sistemler Çağında Mühendisliğin Rolü</h3><p>Bugün mühendislik yalnızca “<strong>çalışan sistemler</strong>” üretmekle ilgili değil.</p><p>Aynı zamanda:</p><p>Karar alan, kendini test eden, davranışını gözlemleyebilen sistemler tasarlamakla ilgili.</p><p>Bu da mühendisliği, teknik bir uğraştan <strong>davranış tasarımı</strong> seviyesine taşıyor.</p><h3>Sonuç: Karar Vermek Yetmez, Davranışı Okuyabilmek Gerekir</h3><p>Yapay zekâ karar vermeye başladığında, artık yalnızca “doğru çalışan” sistemler tasarlamıyoruz.</p><p><strong>Karar alan bir sistem:</strong></p><p><strong>Ortamı yorumlar.</strong></p><p><strong>Sonuç üretir.</strong></p><p><strong>Ve bu sonucun etkisiyle kendi davranışını şekillendirir.</strong></p><p>Bu noktada mühendislik, kod yazmanın ötesine geçer.<br>Davranışı olan sistemleri <strong>anlamaya, izlemeye ve yönlendirmeye</strong> dönüşür.</p><p>Autonomous Agentic Ai, Digital Twin’ler ve Chaos Engineering;<br> farklı teknolojiler gibi görünse de aslında aynı soruya cevap arar:</p><p><strong>Bir sistem karar verdiğinde, bu kararın sonuçlarını nasıl görür, test eder ve taşır?</strong></p><p>Bu yazıda anlatmış olduğum yaklaşım, yapay zekâyı “daha güçlü” yapmaktan çok <strong>daha bilinçli ve daha dayanıklı</strong> hale getirmeyi hedefler.</p><p>Çünkü önümüzdeki dönemde fark yaratacak olan şey..en karmaşık modeli kurmak değil; <strong>karar veren sistemleri yönetebilmektir.</strong></p><h4>Ve belki de asıl fark burada ortaya çıkar:<br><strong> </strong>Karar veren sistemler çağında,<br> en kritik yetkinlik hâlâ <strong>insandadır.</strong></h4><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e85b97f82513" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Kendi Yapay Zeka Asistanımı (Sereth) Nasıl Oluşturdum?]]></title>
            <link>https://medium.com/@eliftutar007/kendi-yapay-zeka-asistan%C4%B1m%C4%B1-sereth-nas%C4%B1l-olu%C5%9Fturdum-0a311b4133c9?source=rss-77dd05013bb8------2</link>
            <guid isPermaLink="false">https://medium.com/p/0a311b4133c9</guid>
            <category><![CDATA[presales]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[ai-assistant]]></category>
            <category><![CDATA[engineering]]></category>
            <dc:creator><![CDATA[Elif TUTAR]]></dc:creator>
            <pubDate>Tue, 12 Aug 2025 13:17:24 GMT</pubDate>
            <atom:updated>2025-08-12T13:17:24.445Z</atom:updated>
            <content:encoded><![CDATA[<p>Part 1’de, yapay zekanın iş ve günlük yaşamda nasıl güçlü bir yardımcı olabileceğini anlattım; verimliliği nasıl artırdığını, karar süreçlerini nasıl hızlandırdığını ve yeni bakış açıları kazandırdığını paylaştım. Şimdi, Part 2’de ise bu vizyonu gerçeğe dönüştürme zamanı. Bu yazıda, kendi yapay zeka asistanım Sereth’i nasıl adım adım kurduğumu, iş hayatıma entegre ettiğimi ve bu süreçte yaşadığım hem profesyonel hem de keyifli anları bulacaksınız.</p><h3>1) İhtiyaçları Belirledim</h3><p>Önce kendime şu soruyu sordum: “Sereth benden ne bekliyor… pardon, ben Sereth’ten ne bekliyorum?” <strong>Cevaplar netti:</strong></p><ul><li>İş süreçlerimi hızlandırmak</li><li>Teknik dokümanları özetlemek</li><li>Ürün karşılaştırmaları yapmak</li><li>Müşterilere özel içerik ve sunum desteği sağlamak</li><li>Toplantı öncesi kısa brifingler hazırlamak</li><li>Rutin işlerde zaman kazandırmak</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cn0Ea0ESiePuwlA-fXFK9w.png" /></figure><h3>2) Doğru Platformu Seçtim</h3><p>Başlangıçta hazır bir GPT platformu kullandım. Bu sayede bir günde çalışır hale geldi. Ancak daha sonra ihtiyaç duyarsam API tabanlı bir sisteme geçebilme opsiyonunu cebimde tuttum. Hız + esneklik = huzur.</p><h3>3) Karakter ve Kuralları Tanımladım</h3><p><strong>Asistanımın kimliğini şekillendirmek hem eğlenceli hem de stratejik bir adımdı:</strong></p><ul><li>Bana “Aethernis efendim” diye hitap ediyor</li><li>Profesyonel ama samimi bir dil kullanıyor</li><li>Gizlilik ve doğruluk ilkelerine uyuyor</li><li>Gerektiğinde hafif esprili, gerektiğinde tam bir mühendis gibi ciddi</li><li>Yanlış bilgi riskinde uyarı veriyor</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kBOV2QgnDWyWVsKj9w0J1w.png" /><figcaption>Talimatlar</figcaption></figure><h3>4) Temel Komutları Belirledim</h3><p><strong>Günlük işlerimi hızlandırmak için kısa ve net komutlar ekledim:</strong></p><ul><li><strong>Özetle:</strong> Belgeleri kısa ve net şekilde özetler</li><li><strong>Karşılaştır: </strong>İki ürünü tablo ile karşılaştırır</li><li><strong>E-posta:</strong> Hazır taslak üretir</li><li><strong>Slayt:</strong> Sunum iskeleti çıkarır</li><li><strong>Risk/Fırsat:</strong> Bir dokümandaki kritik risk ve fırsatları listeler</li><li><strong>Etiketle:</strong> Belgeleri kategorilere ayırır</li><li><strong>Demo Hazırla:</strong> Müşteriye özel ürün demosu akışı çıkarır</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_xVar6yMWD-gE5FLTDOyWg.png" /><figcaption>Temel Komutlar</figcaption></figure><h3>5) Test Ettim ve İş Akışına Dahil Ettim</h3><p><strong>Sereth’i sadece test etmekle kalmadım, günlük hayatımın içine aldım:</strong></p><ul><li><strong>Sunum hazırlığı:</strong> “Bir dakikada şirketinizin operasyonlarını durdurabilecek bir saldırı hayal edin…” gibi etkili açılış cümleleri üretti.</li><li><strong>Müşteri toplantısı:</strong> Toplantı öncesi 5 soruluk olası Q&amp;A listesi hazırladı.</li><li><strong>Müşteriye özel demo: </strong>Her müşteri için sektörüne ve ihtiyaçlarına göre farklı bir demo senaryosu hazırladı.</li><li><strong>Teknik işler: </strong>EPP/EDR/XDR/RMM/Web Protection/Email Protection/CNAPP ürün karşılaştırmalarını 2 dakikada bitirdi.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GWkYjdl1E0LbfJK5wrcRtQ.png" /><figcaption>Teknik işler için örnek;</figcaption></figure><h3>Sonuç)</h3><p><strong>En önemli dersim şu oldu:</strong></p><ul><li>Önce neye ihtiyacınız olduğunu netleştirin, sonra teknolojiyi seçin. Doğru kurulumla, yapay zeka asistanı sadece zaman kazandırmakla kalmıyor; toplantılarda “vay be” dedirten demolar, sunumlar ve fikirlerle işinize renk katıyor. Ve evet, bazen moralinizi yükselten, günü daha keyifli hale getiren bir iş arkadaşına dönüşebiliyor.</li></ul><blockquote><em>Peki Neden </em><strong><em>“Sereth”</em></strong><em> ve </em><strong><em>“Aethernis”</em></strong><em>? Evet, bu soruyu çok duydum. “</em><strong><em>Sereth</em></strong><em>” ismi kulağa hem gizemli hem de teknolojik geliyor; bana güvenilir ama aynı zamanda keskin zekâlı bir asistan hissi veriyor. “</em><strong><em>Aethernis</em></strong><em>” ise… işin eğlenceli kısmı burada başlıyor. Gökyüzü (aether) ve sonsuzluk (infinite) kelimelerinden türettiğim bu isim, hem vizyonumu hem de sınırsız olasılıkları hatırlatıyor. Biraz teknoloji vizyoneri, biraz da sahnede her an sürpriz yapabilecek bir sunumcu yanımı yansıtıyor.</em></blockquote><blockquote><em>Bence bu ikili, iş dünyasında hem sağlam veriye dayalı stratejiyi hem de doğru zamanda gelen yaratıcı bir dokunuşu temsil ediyor. Sonuçta, doğru bilgiyle ikna eder, zekice bir hamleyle hatırlarda kalırsınız.</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0a311b4133c9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Design Your Future: Meet Your AI Assistant]]></title>
            <link>https://medium.com/@eliftutar007/design-your-future-meet-your-ai-assistant-7715ecfdcd02?source=rss-77dd05013bb8------2</link>
            <guid isPermaLink="false">https://medium.com/p/7715ecfdcd02</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-assistant]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[future-technology]]></category>
            <dc:creator><![CDATA[Elif TUTAR]]></dc:creator>
            <pubDate>Sun, 22 Dec 2024 13:50:08 GMT</pubDate>
            <atom:updated>2024-12-22T13:50:08.997Z</atom:updated>
            <content:encoded><![CDATA[<p>Artificial intelligence has become an indispensable part of my life. It has guided me during moments of indecision, helped me create actionable plans to reach my goals, and even reshaped my way of thinking by offering fresh perspectives. From research to navigating challenges in professional life, it has served as a hidden assistant, always by my side.</p><p>One day, I asked myself a question: <strong>“Why not have my own AI assistant, trained with my personal data and tailored to my needs?”</strong> This idea excited me so much that it turned from a mere thought into a real goal.</p><p>In this article, we will explore how AI assistants can simplify our lives and make a tangible difference in various areas. Once you understand the potential of this technology, you’ll be amazed at the possibilities it can unlock!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vWij6QjHpQxGdv02BzEdvQ.png" /><figcaption>Me and My AI Assistant</figcaption></figure><h4><strong>AI: A Powerful Ally in Your Professional Life</strong></h4><p>Artificial intelligence is no longer just a tool; it’s evolving into a reliable ally in the workplace. When used correctly, it can make your time more efficient, accelerate decision-making processes, and simplify task management. In short, it’s a resource that helps you stay ahead in your professional journey.</p><p>Here are some key points to consider:</p><ul><li><strong>AI enhances your skills, not replaces them.</strong> It works alongside you, amplifying your strengths and helping you achieve more.</li><li><strong>It broadens your thinking.</strong> Instead of stifling creativity, AI introduces new knowledge that expands your perspective.</li><li><strong>It’s a reliable guide.</strong> By filtering out misinformation, AI ensures that your decisions are based on accurate data.</li></ul><p>AI opens doors to unknown opportunities and boosts efficiency. See it not as a threat but as an invaluable partner. Don’t hesitate to embrace what lies beyond that door.</p><h4><strong>Ready to Build Your Own AI Assistant?</strong></h4><p>Creating your personal AI assistant might be easier than you think. Here are a few steps to get started:</p><p><strong>Choose the Right Tools:</strong> Platforms like OpenAI GPT API, Dialogflow (Google), and Rasa can be excellent starting points.</p><p><strong>Gather Your Data:</strong> Identify the recurring questions or processes in your work and organize datasets to train your AI assistant.</p><p><strong>Train the Model:</strong> Customize your model to provide the answers you need by using the data you’ve collected.</p><h4><strong>Reimagine Your Workday with AI</strong></h4><p>AI isn’t just useful for technical tasks — it’s also a game-changer in your daily workflow:</p><ul><li><strong>Understanding Customer Needs:</strong> Analyze customer data before meetings to propose tailored solutions.</li><li><strong>Emails and Reporting:</strong> Draft standardized yet personalized email templates and reports effortlessly.</li><li><strong>Time Management:</strong> Optimize your daily schedule to minimize time waste.</li></ul><p><strong>Example:</strong> “I create personalized presentations tailored to my clients’ specific needs, ensuring impactful communication every time.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/948/1*vy-hZQgULg4SEpR1Q8VY9A.png" /><figcaption>Example Content</figcaption></figure><h4><strong>“AI: Your Path to an Easier Tomorrow”</strong></h4><p>Working with AI isn’t just about adopting a tool; it’s about preparing for the future with a reliable companion. By leveraging AI effectively, you can streamline your workflows and develop creative solutions.</p><p>Remember, AI isn’t a competitor. Instead, it’s a helper that learns from you and guides you based on your input. It enhances your skills, helps you utilize information more effectively, and invites you to think from a broader perspective.</p><p>Finally, to those who approach AI with skepticism, I say this: knowledge, curiosity, and an open mind are the keys to understanding its true potential. A desire to learn and explore will unlock growth, both individually and collectively.</p><blockquote>In my next article, I’ll provide a step-by-step guide on how to build your personal AI assistant from scratch. Get ready — by the end of this journey, everyone will have their own digital assistant!</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7715ecfdcd02" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Geleceğinizi Tasarlayın: Yapay Zeka Asistanınızla Tanışma Zamanı]]></title>
            <link>https://medium.com/@eliftutar007/gelece%C4%9Finizi-tasarlay%C4%B1n-yapay-zeka-asistan%C4%B1n%C4%B1zla-tan%C4%B1%C5%9Fma-zaman%C4%B1-e072dc16ce94?source=rss-77dd05013bb8------2</link>
            <guid isPermaLink="false">https://medium.com/p/e072dc16ce94</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai-assistant]]></category>
            <category><![CDATA[yapay-zeka]]></category>
            <dc:creator><![CDATA[Elif TUTAR]]></dc:creator>
            <pubDate>Sun, 22 Dec 2024 13:29:27 GMT</pubDate>
            <atom:updated>2024-12-22T13:30:17.527Z</atom:updated>
            <content:encoded><![CDATA[<p>Yapay zeka hayatımın her köşesinde kendine bir yer edindi. Kararsız kaldığım anlarda bana yol gösterdi, hedeflerime ulaşmak için planlar yapmama yardım etti, hatta bazen yeni bakış açıları sunarak düşünme şeklimi bile değiştirdi. Öyle ki, araştırmalarımda ve iş hayatımdaki zorluklarda adeta gizli bir asistan gibi hep yanımdaydı.</p><p>Bir gün kendime şu soruyu sordum: <strong><em>“Neden bana özel, tamamen benim verilerimle çalışan bir yapay zeka asistanım olmasın?”</em></strong> İşte bu fikir beni öylesine heyecanlandırdı ki, artık sadece hayal değil, bir hedef haline geldi.</p><p>Bu yazıda, yapay zeka asistanlarının hayatımızı nasıl kolaylaştırdığını ve hangi alanlarda bize gerçekten fark yaratabileceğini keşfedeceğiz. Kendi dijital asistanınızı nasıl yaratabileceğinizi öğrenmeye başladığınızda, bu teknolojinin sunduğu olanaklar sizi şaşırtacak!</p><figure><img alt="Me and my AI Assistant." src="https://cdn-images-1.medium.com/max/1024/1*vWij6QjHpQxGdv02BzEdvQ.png" /><figcaption>Me and my AI Assistant.</figcaption></figure><h4>Yapay Zeka: İş Hayatınızda Sizi Destekleyen Güçlü Bir Yardımcı</h4><p>Yapay zeka, sadece bir teknoloji olmanın ötesine geçiyor; iş hayatınızda size destek olan güçlü bir yardımcıya dönüşüyor. Doğru şekilde kullanıldığında, zamanınızı daha verimli hale getirir, karar süreçlerinizi hızlandırır ve işlerinizi daha kolay yönetmenizi sağlar. Kısacası, iş hayatınızda sizi bir adım öne taşıyan bir araç olur.<br>Bu noktada şunu vurgulamak önemli:</p><ul><li>Yapay zeka kişisel becerilerinizin önüne geçmez, aksine onları geliştirir.</li><li>Onunla çalışmak, insan gibi düşünme yeteneğinizi köreltmez; aksine yeni bilgiler ekleyerek zihninizi genişletir.</li><li>Kararlarınızda size güvenilir bir rehber sunar, yanlış bilgilere sapmanızı engeller.</li></ul><p><strong>Yapay zeka, bilinmeyeni keşfetmek ve verimliliği artırmak için bir kapı aralar, ona farklı bir gözle bakın. Ve o kapıdan içeri girmeye korkmayın..</strong></p><h4>Kendi Yapay Zeka’nı Oluşturmaya Hazır mısın?</h4><p>Kendi yapay zeka asistanınızı oluşturmak düşündüğünüzden daha kolay olabilir. İşte bunun için birkaç adım:</p><ul><li><strong>Doğru Araçları Seçin:</strong> OpenAI GPT API, Dialogflow (Google), Rasa gibi platformları kullanabilirsiniz.</li><li><strong>Veri Toplayın:</strong> Sık karşılaştığınız sorular veya iş süreçlerinizle ilgili veri setleri oluşturun. Ve bunları sonrasında yapay zekanızı eğitmek için gruplandırın.</li><li><strong>Modeli Eğitin:</strong> İhtiyacınıza uygun cevapları vermesi için modelinizi özelleştirin. Oluşturduğunuz veri setlerini yapay zekanızı eğitmek için kullanın.</li></ul><h4>Yapay Zeka ile Çalışma Gününüzü Baştan Yaratın!</h4><p>Yapay zeka sadece teknik süreçlerde değil, günlük iş akışında da sizi destekleyecektir:</p><ul><li><strong>Müşteri İhtiyaçlarını Anlama:</strong> Görüşmeden önce müşteriyle ilgili verileri analiz ederek doğru çözümleri önceden belirleyebilirsiniz.</li><li><strong>E-posta ve Raporlama:</strong> Standartlaştırılmış ama kişiselleştirilmiş e-posta taslakları oluşturabilirsiniz.</li><li><strong>Zaman Yönetimi:</strong> Günlük toplantı takviminizi optimize ederek zaman kaybını en aza indirebilirsiniz.</li></ul><p>Örnek: “Ürün anlatımında müşterilerime özel olarak bir sunum hazırlıyorum ve ihtiyaçlarına uygun ürün özelinde anlatım yapıyorum.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qTGqf13NZv5vlCGuUTyG2A.png" /><figcaption>Örnek İçerik</figcaption></figure><h4>“Yapay Zeka ile Yarınlarınız Daha Kolay Olacak”</h4><p>Yapay zeka ile çalışmak, sadece bir araç değil, sizi geleceğe hazırlayan bir yol arkadaşıdır. Onu doğru kullanarak iş süreçlerinizi daha verimli hale getirebilir ve yaratıcı çözümler geliştirebilirsiniz.<br>Unutmayın, yapay zeka bir rakip değil; onu nasıl eğitirseniz size o şekilde rehberlik eden bir yardımcıdır. Sizin becerilerinizi geliştirir, bilgiyi daha etkili kullanmanıza yardımcı olur ve sizi daha geniş bir perspektiften düşünmeye davet eder.</p><p>Son olarak, yapay zeka hakkında farklı görüşlere sahip olanlara şunu hatırlatmak isterim: Bilgi, araştırma ve açık fikirli bir yaklaşımla derinleşir. Bu konudaki merak ve öğrenme isteği, hem bireysel hem de kolektif gelişimin anahtarıdır.</p><blockquote>Bir sonraki yazımda, kendi yapay zeka asistanınızı sıfırdan nasıl oluşturabileceğinizi adım adım anlatacağım. Hazır olun, çünkü bu yolculuğun sonunda herkesin kendi dijital asistanı olacak!</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e072dc16ce94" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Enhance Your SAST Tool with Retrieval-Augmented Generation?]]></title>
            <link>https://medium.com/@eliftutar007/how-to-enhance-your-sast-tool-with-retrieval-augmented-generation-8eff3dc40355?source=rss-77dd05013bb8------2</link>
            <guid isPermaLink="false">https://medium.com/p/8eff3dc40355</guid>
            <dc:creator><![CDATA[Elif TUTAR]]></dc:creator>
            <pubDate>Wed, 25 Sep 2024 11:56:57 GMT</pubDate>
            <atom:updated>2024-09-25T11:56:57.151Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>The idea for this blog post came from discussing RAG in my “Prompt Engineering” course at Bahçeşehir University. As someone who enjoys integrating different topics, I developed this idea while working as an AppSec researcher , leading me to write “How to Enhance Your SAST Tool with Retrieval-Augmented Generation.” Now, let’s uncover how we can revolutionize SAST tools with this technology!</blockquote><p>In my previous blog post, ‘<a href="https://medium.com/@eliftutar007/software-development-life-cycle-sast-and-sca-combined-225f94749a5f">Software Development Life Cycle: SAST and SCA Combined</a>,’ I highlighted the importance of combining Static Application Security Testing (SAST) and Software Composition Analysis (SCA) of SAST &amp; SCA in the software development lifecycle. Now, I’m writing this blog post because I believe that adapting Retrieval-Augmented Generation (RAG) to new world technologies can elevate the SAST tool to the next level.</p><p>In the rapidly evolving landscape of software development, ensuring the robustness and security of applications from the earliest stages has never been more critical. Among the range of tools at our disposal, SAST stands out for its ability to scrutinize source code for potential security vulnerabilities without requiring program execution. However, as applications grow in complexity and the variety of potential threats expands, traditional SAST methodologies are challenged to keep pace. It’s here that the fusion of SAST with cutting-edge technologies like RAG and generative AI, specifically Large Language Models (LLMs), promises to revolutionize our approach to securing our digital infrastructures.</p><p>In this article, I explore how the integration of RAG into SAST tools can significantly enhance the effectiveness of static application security testing. I’ll delve into an understanding of RAG and its synergy with machine learning and generative AI to bolster the capabilities of traditional SAST scans. Additionally, I’ll outline the benefits of this integration, ranging from enhanced detection rates of security vulnerabilities to refined recommendations for remediation. Following that, I’ll provide a step-by-step guide on implementing RAG within SAST tools, addressing common challenges, and offering solutions to ensure a smooth integration process. By the end of this discussion, I aim to provide a comprehensive roadmap to leveraging the power of RAG to enhance SAST’s efficiency, making it a formidable tool in our cybersecurity arsenal.</p><h3>What is Retrieval-Augmented Generation (RAG)?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4dS8-oAFuXBg6crYMuNdJQ.jpeg" /><figcaption>How to RAG works</figcaption></figure><p>Retrieval Augmented Generation (RAG) is a technique designed to enhance language models (LLMs) by providing additional context, allowing them to generate better and more specific responses.</p><p><strong>Context Enhancement:</strong> LLMs are trained on publicly available data and are highly intelligent, but they often lack the specific context needed to answer certain queries accurately. RAG addresses this by supplying the necessary context to the LLMs, enabling them to provide more accurate and relevant answers.</p><p><strong>Dynamic Knowledge Integration:</strong> RAG allows for the insertion of new knowledge or capabilities into LLMs without permanent changes. This means the model can access and utilize up-to-date information dynamically without needing to be retrained.</p><p><strong>Combining Retrieval and Generation: </strong>RAG merges the retrieval of relevant text with the answer generation process. By integrating retrieved information into the generation process, RAG improves the performance of LLMs, yielding more precise answers than what is possible with pre-training alone.</p><p><strong>Efficient Information Retrieval:</strong> RAG is particularly useful in scenarios where the information exceeds the processing capacity of a standard LLM context window. For example, when dealing with a large book, RAG breaks it into manageable segments, creates embeddings for each segment, and matches the embeddings of the query with those segments. The vector database then retrieves the closest matches, providing accurate and efficient information retrieval that can be fed into the LLM’s context window.</p><p>In summary, RAG is a powerful technique that enhances the capabilities of language models by providing them with the necessary context and dynamically integrating new knowledge, resulting in more accurate and contextually relevant responses.</p><h3>How RAG Functions</h3><ol><li><strong>Initial Analysis</strong>: RAG starts by analyzing the user’s input to understand the intent and specific information requirements. This crucial step determines the direction of the subsequent data retrieval.</li><li><strong>Data Retrieval</strong>: Utilizing advanced algorithms, RAG searches and retrieves information from external databases, APIs, or document repositories relevant to the query.</li><li><strong>Data Transformation</strong>: The retrieved data is converted into numerical vectors, making it manageable for the AI to perform complex relevance calculations.</li><li><strong>Content Augmentation</strong>: The AI model then integrates this external data with the original user prompt, enhancing the generated content’s relevance and accuracy.</li></ol><h3><strong>Advantages of RAG</strong></h3><ul><li><strong>Increased Precision and Contextual Relevance:</strong> By pulling in information from external sources, RAG enhances the relevance and accuracy of generated content, ensuring it’s based on reliable and up-to-date knowledge.</li><li><strong>Enhanced Contextual Consistency:</strong> The retrieval step provides a solid foundation for generating responses, leading to more coherent and contextually grounded outputs.</li><li><strong>Optimized Resource Usage:</strong> RAG models leverage pre-existing structured data sources, reducing the need for generating everything from scratch, which lowers computational costs and improves efficiency.</li><li><strong>Versatility Across Modalities:</strong> RAG’s adaptability allows it to function across different formats (e.g., text, image) and domains, making it useful for a wide range of tasks, from content creation to summarization.</li></ul><p>By integrating these strengths, RAG elevates AI performance, especially in tasks where contextual accuracy and relevance are critical.</p><h3><strong>Overview of SAST</strong></h3><p>Static Application Security Testing (SAST) is a critical methodology in the field of application security that focuses on examining the inner workings of an application without executing it. By inspecting the source code, bytecode, or binaries, SAST identifies security vulnerabilities early in the software development lifecycle (SDLC). This proactive approach allows developers to fix issues before they escalate into serious security threats, ensuring the application is secure when it goes live.</p><p>SAST operates by analyzing code against predefined security rules and best practices, which are tailored to identify common security flaws such as SQL injection, Cross-Site Scripting (XSS), and buffer overflows. These vulnerabilities, if not addressed, can enable attackers to compromise the application by executing malicious code, accessing sensitive data, or escalating privileges.</p><h4><strong>Key Techniques in SAST:</strong></h4><ol><li><strong>Data Flow Analysis:</strong> Tracks the flow of data within the application to identify how it moves and transforms, pinpointing insecure practices that might expose vulnerabilities.</li><li><strong>Taint Analysis:</strong> Detects untrusted or unsanitized data flows, particularly user inputs, that could lead to potential security breaches if not properly validated or sanitized.</li></ol><h4><strong>The Importance of SAST in Modern Development</strong></h4><p>With the growing use of artificial intelligence in code development, new security risks have emerged, making the use of SAST tools more critical than ever. These tools are essential for identifying and addressing security vulnerabilities, particularly those introduced by AI-generated code. By shifting left in the software development lifecycle (SDLC), developers are expected to address security issues early on, and SAST tools provide the necessary support to detect and resolve vulnerabilities at every stage of development.</p><h3><strong>Benefits of Integrating RAG into SAST Tools</strong></h3><figure><img alt="This image was generated by ChatGPT for my blog post." src="https://cdn-images-1.medium.com/max/1007/1*RO4Cv2YX1Vm7WmNt2K7Lnw.jpeg" /></figure><p><strong>Improved Threat Detection:</strong><br>RAG integration within SAST tools greatly enhances the detection of complex security threats by utilizing advanced threat modeling techniques. This is especially valuable in cloud-based systems, where the attack surface is broader due to the intricate architecture. Leveraging frameworks like the STRIDE methodology, RAG helps security teams develop comprehensive strategies to address these challenges, resulting in a more thorough detection of potential vulnerabilities.</p><p><strong>Enhanced Code Analysis:</strong><br>The combination of RAG with SAST tools harnesses the power of generative AI to deliver real-time, AI-generated fixes during code scans. This not only accelerates the remediation process but also improves the precision of code analysis. Developers receive actionable, specific suggestions for resolving vulnerabilities directly within their pull requests, making the process of securing code much more efficient and ensuring that vulnerabilities are addressed early in the development cycle.</p><p><strong>Real-Time Updates and Adaptation:</strong><br>By integrating real-time data into the SAST process, RAG allows for continuous updates that align with the evolving threat landscape. This dynamic capability ensures that security measures stay relevant and responsive to the latest risks. As RAG-enhanced SAST tools adapt to new vulnerabilities, they provide more accurate and contextually relevant security assessments, supporting a proactive security strategy that reduces the likelihood of exposure to emerging threats.</p><h3>Implementation Steps for Integrating RAG with SAST</h3><h4>1. Setting Up the RAG Pipeline</h4><p><strong>Initialize Retrieval Sources:</strong><br>Begin by configuring internal and external data sources that will provide critical security information. These can include public repositories like <strong>GitHub</strong>, vulnerability databases such as <strong>NVD</strong> and <strong>CVE</strong>, and security guidelines like <strong>OWASP Top 10</strong> and <strong>SANS CWE Top 25</strong>. In addition, real-time threat intelligence feeds from services like <strong>AlienVault OTX</strong> and security communities like <strong>Stack Overflow</strong> ensure the retrieval system is constantly updated.</p><p><strong>Embedding and Vector Database Configuration:</strong><br>Set up vector databases such as <strong>Pinecone</strong> or <strong>FAISS</strong> to store document embeddings. These embeddings are vital for matching queries to relevant documents based on semantic similarity, enabling efficient retrieval. <strong>ElasticSearch</strong> and <strong>Lucene</strong> can also be integrated for improved indexing and search capabilities.</p><p><strong>Integration with Generative Models:</strong><br>Integrate the retrieval system with a generative AI model like <strong>GPT-4</strong> or <strong>Hugging Face Transformers</strong>. These models utilize the retrieved information to generate contextually accurate content, such as remediation suggestions for vulnerabilities detected in the code. Tools like <strong>GitHub Copilot</strong> can further streamline the process by integrating directly into the development environment and offering real-time feedback on code.</p><h4><strong>2. Performing Tests and Evaluations</strong></h4><p><strong>Initial Testing:</strong><br>After the integration is complete, run initial tests to verify the functionality of the RAG-enhanced SAST tool. This involves analyzing different codebases to ensure the retrieved data is being used effectively by the generative model. For example, checking whether vulnerability data from sources like <strong>CVE</strong> is accurately influencing the suggestions provided by the system.</p><p><strong>Evaluation of Effectiveness:</strong><br>Metrics such as the <strong>accuracy of vulnerability detection</strong>, reduction in <strong>false positives</strong>, and <strong>remediation time</strong> should be used to assess the effectiveness of the RAG integration. A higher detection rate with fewer false alarms will indicate that the system is performing optimally. At this stage, it’s critical to compare RAG-enhanced scans with traditional SAST tools to quantify improvements in precision and response times.</p><p><strong>Continuous Improvement:</strong><br>Based on the outcomes of these evaluations, refine both the RAG setup and SAST configurations. This can involve tweaking the retrieval sources, adjusting the generative model’s behavior, or even introducing new data sources. Continuous monitoring and updates ensure that the tool remains aligned with the latest security threats and best practices, maintaining its efficiency in identifying and remediating vulnerabilities.</p><p>This extended version includes both the setup and iterative improvement process, emphasizing that continuous testing and adaptation are key to successful RAG and SAST integration.</p><h4>Conclusion</h4><blockquote>Integrating Retrieval-Augmented Generation (RAG) into Static Application Security Testing (SAST) tools marks a transformative step in enhancing the security and efficiency of modern software development. By leveraging RAG’s ability to deliver real-time, contextually relevant insights, we can detect and remediate vulnerabilities more accurately and effectively, ensuring proactive protection throughout the development lifecycle.</blockquote><blockquote>While the challenges of integrating RAG with SAST tools reveal the complexity of the process, they also highlight the potential for significant advancements in securing our digital infrastructure against sophisticated threats. As we continue to push the boundaries of AI and machine learning in cybersecurity, we move toward a future where software development is inherently more secure, reliable, and resilient.</blockquote><p>#Sast #Rag #Sdlc #Appsec</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8eff3dc40355" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Software Development Life Cycle: SAST and SCA Combined]]></title>
            <link>https://medium.com/@eliftutar007/software-development-life-cycle-sast-and-sca-combined-225f94749a5f?source=rss-77dd05013bb8------2</link>
            <guid isPermaLink="false">https://medium.com/p/225f94749a5f</guid>
            <category><![CDATA[sast]]></category>
            <category><![CDATA[appsec]]></category>
            <category><![CDATA[ca]]></category>
            <category><![CDATA[sdlc]]></category>
            <category><![CDATA[application-security]]></category>
            <dc:creator><![CDATA[Elif TUTAR]]></dc:creator>
            <pubDate>Thu, 18 Jan 2024 07:03:16 GMT</pubDate>
            <atom:updated>2024-01-18T07:03:16.055Z</atom:updated>
            <content:encoded><![CDATA[<blockquote><em>Let’s dive into how tiny details in software development can lead to big security issues, and how using SAST and SCA together can strengthen your code’s security. We’ll see why blending these tools within the Software Development Life Cycle (SDLC) is great for making safer software.</em></blockquote><blockquote><em>Author: Elif Tutar — AppSec Researcher aka JigglyPuff ✨</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*CJkovcYaU78sFV54.png" /></figure><p>As practitioners in the field of secure software development, we are often equipped with tools like SAST and SCA. Imagine the compounded benefits if we could merge these tools’ functionalities into one seamless product.</p><p>In this discussion, we will break down the distinct roles of Software Composition Analysis and Static Application Security Testing.</p><h3>SCA — Software Composition Analysis</h3><p>SCA is about keeping your software safe by keeping an eye on the external bits of code you use, like open-source components, and their licenses. It checks your software to find and manage these parts, making sure they’re safe and legal to use.</p><p>Big security incidents like SolarWinds show why SCA is so important. Hackers often hit the weakest part, like old or hidden code that gets into your software without you realizing it. SCA helps avoid this by using wide-ranging sources for security checks, not just the standard databases. It’s become a key tool for fighting cyber threats, especially with so much open-source code in use today.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/886/0*srFV6qU7WNM4udTE.png" /><figcaption>It’s very dangerous to open a package without knowing what’s inside, especially if you don’t know where it came from</figcaption></figure><p><strong>Open Source License Compliance</strong>: SCA tools facilitate compliance with open-source licenses by reporting a detailed list of licenses associated with dependencies. Strong policy engines within SCA tools can also be configured to block builds if an out-of-policy license is detected.</p><p><strong>SBOM Generation</strong>: SCA tools play a crucial role in generating Software Bill of Materials (SBOMs), providing a detailed inventory of software components.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vkKvqSqcWPZEmnwb.png" /><figcaption>dependency graph example from codethreat platform</figcaption></figure><p><strong>Open Source Vulnerability Management</strong>: Beyond creating an inventory of known vulnerabilities (CVEs), SCA aids in prioritization and remediation. For instance, tools like NIST National Vulnerability Database provide CVSS and EPSS scores for vulnerabilities to assist in prioritization. Moreover, they offer insights into the fix and code path, streamlining the remediation process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*io__8A4tLOCRN_Jl.png" /><figcaption>example sca output</figcaption></figure><blockquote><em>In summary, Software Composition Analysis is an integral part of the software development process, enhancing security by ensuring comprehensive identification, tracking, and management of opensource components and potential vulnerabilities</em></blockquote><h3>SAST — Static Application Security Testing</h3><p>Static Application Security Testing is an important technique used to check the building blocks of an application — like its source code or the ready-to-run code — for any weak spots that could be a security risk. This check is done without running the application. It’s usually done early on when the application is being created, so that developers and security experts can fix these issues before the application is finished and used by others.</p><p>SAST operates by identifying potential vulnerabilities through predefined rulesets or security patterns rooted in established security best practices. These rulesets are crafted based on common coding errors and known security pitfalls, including issues such as SQL injection, cross-site scripting (XSS), buffer overflows, and other vulnerabilities that could potentially enable threat actors to escalate privileges, access restricted data, or execute malicious code.</p><h3>SAST Techniques</h3><p><strong>Data Flow Analysis:</strong> This method tracks how data traverses through the application. It encompasses control flow analysis to understand the execution order of different code sections and code path analysis to comprehend conditional statements, loops, and varied execution paths.</p><p><strong>Taint Analysis:</strong> Focused on tracing the flow of untrusted or tainted data, this technique helps identify potential security risks arising from improperly sanitized or validated user input.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*FcQZvUAfjfhiOdn0" /><figcaption>points-to (pointer) analysis decision making</figcaption></figure><blockquote><em>SAST Capabilities:<br>• Pinpoint Flaws in Proprietary Code<br>• Detect Weaknesses Early in the SDLC, Reducing Costs<br>• Identify Weaknesses Before Code Deployment into Production</em></blockquote><p>A SAST tool is the embodiment of this methodology, focusing on addressing security issues within an organization’s proprietary software. By scanning the source code for known vulnerable code patterns, SAST tools generate insights into potential security flaws and vulnerabilities. Operating early in the SDLC, these tools provide real-time feedback to developers during coding, facilitating issue remediation before code progression to subsequent phases of the SDLC.</p><p><strong>SDLC process is integral to SCA and SAST. What does SDLC stand for?</strong></p><p>The Software Development Life Cycle (SDLC) encompasses the systematic process of planning, designing, implementing, testing, deploying, and maintaining a software project. This process ensures the organized management of software from its inception to completion</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/470/0*G6gR87vJmF4XUllj.png" /><figcaption>a brief code repository security operations</figcaption></figure><p><strong>The Role of SCA and SAST in SDLC:</strong></p><p>Software Composition Analysis and Static Application Security Testing play vital roles in the software development lifecycle.</p><p>SCA follows the “shift-left” principle, becoming an integral part of the early development stages. Here, it offers ongoing feedback to developers, informing them about the security and licensing status of open-source components. This early integration allows for the timely identification of potential risks and provides developers with opportunities to address issues promptly. With the “shift-everywhere” approach, SCA’s influence extends beyond development, reaching testing, integration, and operational stages. Throughout the software life cycle, SCA remains vigilant, continuously monitoring and managing the security status of open-source components.</p><blockquote><em>Similarly, SAST is heavily employed in the initial phases of development, focusing on code writing and compilation. During these stages, SAST identifies possible security vulnerabilities, offering developers immediate insights for intervention. Adopting the “shift-everywhere” perspective, SAST goes beyond development, actively participating in testing, integration, and operational stages. Its goal is to provide ongoing monitoring and resolution of security vulnerabilities throughout the entire lifespan of the application.</em></blockquote><p><strong>Shift-Everywhere Approach:</strong></p><p>In both cases, the “shift-everywhere” approach expands the effectiveness of SCA and SAST beyond the early stages of development, integrating them into testing, integration, and operational stages. This approach promotes the widespread adoption of security considerations throughout every stage of the software life cycle, fostering the development of more secure software practices.</p><p><strong>Comprehensive Security Analysis:</strong></p><ul><li>SCA identifies security vulnerabilities in third-party components.</li><li>SAST reveals security vulnerabilities in the source code through code analysis.</li><li>When used together, they enable a comprehensive analysis of both external dependencies and the security aspects of original code.</li></ul><p><strong>Advanced Threat Detection:</strong></p><ul><li>SCA identifies security threats in known third-party components.</li><li>SAST detects unknown vulnerabilities at the source code level.</li><li>When employed together, they provide a broader scope against both known and unknown security threats.</li></ul><p><strong>Remediation and Implementation Speed:</strong></p><ul><li>SCA facilitates the rapid updating of libraries with identified security vulnerabilities.</li><li>SAST, pinpointing errors at the source code level, provides developers with early warnings.</li><li>When used together, these two methods enable the early detection and correction of errors, resulting in a faster and more effective security response.</li></ul><h3>Conclusion</h3><blockquote><em>while using AppSec tools like SCA and SAST is crucial for robust software security, it’s true that they can be challenging to maintain in real-life scenarios. The key is to integrate these tools in a way that aligns with and simplifies your organization’s existing practices, ensuring both effectiveness and ease of implementation. Balancing security needs with practicality is essential in today’s dynamic tech environment.</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=225f94749a5f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>