<?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 Nikola Shekerev on Medium]]></title>
        <description><![CDATA[Stories by Nikola Shekerev on Medium]]></description>
        <link>https://medium.com/@nikola-shekerev?source=rss-d82a98940996------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*saQp1iDqx8qUP7AaI3yHpw.jpeg</url>
            <title>Stories by Nikola Shekerev on Medium</title>
            <link>https://medium.com/@nikola-shekerev?source=rss-d82a98940996------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 11:51:04 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nikola-shekerev/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[The start of Hardware as a Service]]></title>
            <link>https://nikola-shekerev.medium.com/the-start-of-hardware-as-a-service-74e04404aefd?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/74e04404aefd</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[gpu]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Thu, 12 Nov 2026 18:01:27 GMT</pubDate>
            <atom:updated>2026-11-16T06:31:24.788Z</atom:updated>
            <content:encoded><![CDATA[<p>The CEO of NVIDIA screwed up. But he is also the only techbro oligarch who:<br>- supported the existence of open models<br>- spoke openly (example — CBS interview) how the narrative of rogue AI-s has the objective of loosening legislation, so Big Tech is not liable — “AI did this bad thing, not us”<br>- tried to negotiate relations with China that are not a kill-race</p><p>He screwed up. He is not a “good guy”. In all likelihood, he is a terrible human. But all this is also just a pretext to get rid of someone politically inconvenient : )</p><p>Chances are his successor will fall in line with the push for gpu-s having hardware locks on what can and cannot run on them.</p><p>Here is a simple comparison. Most of us consider Microsoft a terrible monopolist. Imagine how much worse Microsoft would be, if ~25 years ago they have succeeded in their efforts to ban Linux.</p><p>Welcome to “Hardware as a service”</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=74e04404aefd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[On gambling that targets children]]></title>
            <link>https://nikola-shekerev.medium.com/on-gambling-that-targets-children-19a8d0f1c239?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/19a8d0f1c239</guid>
            <category><![CDATA[children]]></category>
            <category><![CDATA[gambling]]></category>
            <category><![CDATA[gaming]]></category>
            <category><![CDATA[parenting]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Sat, 24 Oct 2026 14:08:54 GMT</pubDate>
            <atom:updated>2026-10-24T14:08:54.204Z</atom:updated>
            <content:encoded><![CDATA[<p>This is a personal real story</p><p>I like videogames, so I experiment with different genres. A portion of this market operates with microtransactions, aka micro payments. A considerable part of those products employ strategies quite similar to gambling. Mostly the difference is — you bet real money or a direct in-game derivative of real money and get something in-game, not money.</p><p>In the context of parenting, this matters because<br>- A significant portion of these games are for mobile — the thing in the pocket of your kids<br>- Micro payments tend to skip by parent supervision</p><p>I like videogames, so I tried some of those. 32 years old at the time. As a responsible adult, I decided that it is fine to indulge in a bit of spending. I spent around 1500 dollars in the course of a few months, before I caught myself.</p><p>I carefully curate what I play. The games that ate that money were competently made and enjoyable. But a lot of that comes from my curation. I picked games that are least likely to be poison. If a parent is doing the curation, this is the best case scenario. I got something in return from these games, but the point is that what I got in return was not somehow special or unique. I could have gotten that from somewhere else for x50 times less money. So it is scam, there is literally no reason to do that. And the vast majority of targets are kids</p><p>If you are a parent, find a way to block games (or any process) from accessing your credit card. That is a sufficient fix. The games industry has thousands of magnificently amazing games that are almost free.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=19a8d0f1c239" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your business no longer owns your proprietary data and IP]]></title>
            <link>https://nikola-shekerev.medium.com/your-business-no-longer-owns-your-proprietary-data-and-ip-2c4873984fce?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/2c4873984fce</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[security]]></category>
            <category><![CDATA[data-engineering]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Fri, 09 Oct 2026 18:34:47 GMT</pubDate>
            <atom:updated>2026-11-16T06:34:35.890Z</atom:updated>
            <content:encoded><![CDATA[<p>INTRODUCTION</p><p>In my previous article, <a href="https://nikola-shekerev.medium.com/how-you-pay-for-ai-0ca4bbe13063">linked here</a>, I said that for the last few years, viable pathways have emerged for the proprietary data and IPs of our businesses to be stolen, with ZERO risk for the perpetrators. Those pathways are obvious, but I outline one here, just for completeness.</p><p>WHO IS THE LIKELY PERPETRATOR</p><p>Everyone with some access to your stuff and at least some technical competency can pull this off. This includes most of your employees and all vendors your business depends on. But the companies that will gain the most with minimal effort and zero risk are cloud providers — Amazon, Google, etc.</p><p>So I assume it would be them.</p><p>HOW WILL IT HAPPEN</p><p>Imagine you are a C-level executive or founder, in a Cloud Provider, or another tech giant. You have access to the proprietary data and IP of your customers. You want to exploit that data, but the contract forbids you. The following steps outline how you can still do it, with zero risk for you or your business.</p><p>AI AS THE MAIN TOOL</p><p>You will use AI to reverse engineer and exfiltrate the business know-how of that corpus of proprietary data. The AI will produce some synthetic artifact. Then you will use AI, again, to anonymize the output. Since the anonymization happens AFTER the exfiltration, it does not protect the original owner of the data. The purpose of the anonymization is to make it harder for the artifact to be eventually traced back to you.</p><p>PART 1 — NO HUMAN IN THE LOOP</p><p>AI = NO HUMAN IN THE LOOP = MAIN DEFENSE</p><p>The main weakpoint of any scheme is humans — gossipers, witnesses, leakers, whistleblowers, the discontent. AI means no human in the loop. AI solves the main weakpoint of any data crime. Only the perpetrator knows what runs in the contract-violating part of this process and is the main beneficiary. If there is not a single other person who can testify, the perpetrator is not just going to self-incriminate themselves in a court of law.</p><p>Legal representatives are not tech experts. Experts may be used in an advisory role, sure. But no other human testimony, at all, completely changes the game of all such cases.</p><p>TECHNICALITIES</p><p>I speak of A MOUNTAIN of work, but the level of complexity of that work is manageable enough to be done by AI</p><p>Tech giants can host frontier models inside their own network, so there is no weakspot/leak there.</p><p>The produced synthetic artifact can be used internally and/or sold to data brokers.</p><p>For the first few hundred iterations of such exfiltration, the artifacts may not be great, until the pipeline is perfected, but that is irrelevant. A lot of that improvement work can be somewhat automated, again, with AI.</p><p>Ideally, the artifact will not be accessible to humans, but in the unlikely event that it did, the company can claim it produced it internally from their own business know-how with AI assistance. Synthetic artifacts are common.</p><p>If caught, another path of plausible deniability is to blame the AI going rogue. The whole narrative of rogue AI-s (accidentally or not) functions as a liability offload</p><p>BUT THIS IS NOT LEGAL</p><p>If a crime is so invisible, that there is zero way for it to be proven, then the law cannot be applied. If a law cannot and will not be applied, said law does not exist.</p><p>This is inevitable, not because it is smart or legal. This is inevitable, because, unless a perpetrator is so catastrophically incompetent as to self-incriminate, AI removes all legal risk here.</p><p>No legal risk does not make something legal, but in practical terms, that difference is irrelevant.</p><p>WHO ARE THE TARGETS</p><p>The targets will not be apex predators like Apple, who can hit back hard. The targets will be small businesses, like you and me.</p><p>BUT MY SECURITY IS PERFECT</p><p>Let’s assume your security is perfect (not likely). You still depend on all vendors you use to NOT break their part of the contract. But now, there is no risk for them, if they do.</p><p>THIS IS NOT THE FIRST TIME</p><p>Ages ago, even without the massive advantage of AI, Amazon pulled this off in their retail business, and got away with it when caught.</p><p>PART 2 — DERIVATIVES</p><p>I give an AI some content and ask it to give it back to me in some processed form. That is a derivative. Above I talk about summaries, but this article is valid for any derivative</p><p>Derivatives are currently in a legal grey area, because of how politically volatile the topic is — models were trained on torrents so if I ask it to re-make the final season of Game of Thrones, who is the owner — the prompter, the model owners, the books author or the studio holding the rights? Since trillions of dollars of investments are on the line here (meaning — political backing), legislators avoid strong positions.</p><p>AI performs exceptionally well in producing a specific type of derivative — retell this data source with your own words and names, so a reader cannot trace it back to the original.</p><p>TRAINING ON DERIVATIVES WILL BE LEGAL IN THE FUTURE</p><p>So derivatives are legally ambiguous and legislation there — politically volatile.</p><p>If we go back in 2022 and suggest that AI would be trained on petabytes upon petabytes of torrents, people would have assumed this would be illegal and not viable. Well, now we know this is exactly how AI is trained. The court of law said it is fine. Here I describe the process of exfiltrating proprietary data and IP via AI derivatives. So in 2030, this process will be treated the same way as training on torrents — it will be normalized and fine.</p><p>DERIVATIVES OBJECTIVE</p><p>A human can clearly see that a summary of some work is derived from the work. The objective of a derivative is not to deceive the human. Derivatives are NOT deception.</p><p>The objective of a derivative is to add ambiguity in order to be less valuable as evidence. The derivative achieves this objective by sprinkling enough noise to make it unclear which is the exact point of origin of the data — is it this specific database or this specific repository.</p><p>Think of derivatives as a social engineering hacking attack that does NOT target the software but the legal system instead</p><p>DERIVATIVES ARE TRIVIAL TO MAKE</p><p>Since the objective of the derivative is not to deceive, it is trivial to make. The perpetrator does NOT ask the AI to think or be strategic, just to add variance and noise. This is trivial even for AI</p><p>DERIVATIVES ON DEMAND</p><p>The perpetrator may keep a backup of the original data. So if a buyer wants information of type Z on legally protected entity X, the perpetrator will not sell “Company X has Z1, Z2, Z3” but generate dynamically for that trade something like “Companies that have publicly visible traits X1, X2, X3 seemingly tend to have non-public traits Z1, Z2, Z3”. Since derivatives can be generated on demand, the perpetrator can use this to his advantage and generate something that maximizes what the buyer wants and also hide his tracks, for the specific deal.</p><p>SUMMARY</p><p>The point of this article is that AI introduces 2 new types of social engineering hacking attacks</p><ul><li>No human in the loop means there are no accomplices in the scam who may leak or testify</li><li>Derivatives exploit the human limitations of the legal system</li></ul><p>Together, these 2 attack vectors remove all the legal risk of data crimes</p><p>SIDE NOTES</p><p>The following notes are clarifications on recurring responses I got from readers.</p><p>SIDE NOTE: BUT THIS IS TOO COMPLEX</p><p>No, it is not. The blocker in the past was not complexity. The blocker was that such schemes required a mountain of work, meaning humans and leaks and overwhelming legal risk. AI removed all that risk. There is no risk.</p><p>SIDE NOTE: CLOUD PROVIDERS DO NOT HAVE THAT ACCESS</p><p>There is this misconception that Cloud Providers do not have the access I speak of. Imagine the following scenario — I have physicals access to a bare metal server, the actual hardware, and I have admin access to its boot, OS, disk , and most importantly — admin access to the virtualization engine running on it. But I do not have access to a vm that runs on it. Sure, that access does not come out of the box. But I have everything I need to hack it.</p><p>SIDE NOTE: NOT ABOUT CODE</p><p>In this article, I do not talk about exfiltrating your code. No one cares about your code. I talk about the know-how of your business.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2c4873984fce" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How you pay for AI]]></title>
            <link>https://nikola-shekerev.medium.com/how-you-pay-for-ai-0ca4bbe13063?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/0ca4bbe13063</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[security]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Fri, 09 Oct 2026 12:30:48 GMT</pubDate>
            <atom:updated>2026-10-12T07:41:38.590Z</atom:updated>
            <content:encoded><![CDATA[<p>SUMMARY</p><p>The point of this article is that the risks of AI are not technical. The tech is fine. The problem are humans, who use it as a weapon against customers and small businesses</p><p>IDEA : PRIMARY CURRENCY IS ATTENTION</p><p>As a paying customer, my primary currency is my attention.</p><p>AI has become a signal of low effort, and, consequently, bad service. Low effort is not worth the primary currency of most paying customers.</p><p>Side notes:</p><ul><li>Money matters, but your primary currency is still you attention</li><li>“signal” means not always, but too often</li></ul><p>IDEA : I WANT TO OWN MY STUFF</p><p>AI companies scraped the whole internet. Then they trained on petabytes of torrents. The internet is a lot of AI now, meaning less valuable for training. The next target is the proprietary data and intellectual property of your business. Not likely, but let’s assume your security is perfect. For the last 2 years, frontier models have improved a lot. This means that if a vendor has access to your stuff, AI now gives them multiple pathways, not just one, where they can exfiltrate your stuff without you having any way to prove that. Vendors can break their contracts without risk now.</p><p>If a crime can profit someone tremendously, and the crime cannot be proven, there is no risk, so of course it will happen.</p><p>The likely perpetrators will be cloud providers — Amazon, Google. If your security is perfect (not likely), you still depend on vendors to not break their part of the deal.</p><p>The targets will not be apex predators like Apple, who can hit back hard. The targets will be small businesses.</p><p>How this would happen is obvious, but here is <a href="https://nikola-shekerev.medium.com/your-business-no-longer-owns-your-proprietary-data-and-ip-2c4873984fce">an example</a>, for completeness</p><p>This is not about law or ethics. I just want to own my stuff. Chances are, you want that too.</p><p>IDEA : AI IS THE PERFECT TOOL FOR WEAPONIZED INCOMPETENCE</p><p>The reasoning is non-trivial, so in separate article here: <a href="https://medium.com/@nikola-shekerev/the-unexpected-connection-between-ussr-and-ai-4c25aed2c6f3">https://medium.com/@nikola-shekerev/the-unexpected-connection-between-ussr-and-ai-4c25aed2c6f3</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0ca4bbe13063" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Facebook works as a paywall backdoor]]></title>
            <link>https://nikola-shekerev.medium.com/facebook-works-as-a-paywall-backdoor-82bca71334b3?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/82bca71334b3</guid>
            <category><![CDATA[security]]></category>
            <category><![CDATA[paywall]]></category>
            <category><![CDATA[news]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Thu, 17 Sep 2026 08:22:56 GMT</pubDate>
            <atom:updated>2026-09-17T08:29:31.549Z</atom:updated>
            <content:encoded><![CDATA[<p>Here is how it works:</p><ul><li>You open a link to Financial Times</li><li>You see it requires a subscription, you cannot just read the article</li><li>You post the link to Facebook instead</li><li>You open the link from your own post, and you can read it without paying</li></ul><p>This works with other social networks, for example works with Linkedin</p><p>This works for other paywalled sites, not just Financial Times.</p><p>This does not break Terms of Service, afaik social networks pay to popular outlets for traffic.</p><p>This is not a hack, just advertisement with extra steps. But it is funny that they do it : )</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=82bca71334b3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why social networks are unhealthy, in short]]></title>
            <link>https://nikola-shekerev.medium.com/why-social-networks-suck-in-short-42240e31fd47?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/42240e31fd47</guid>
            <category><![CDATA[social-media]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Fri, 07 Aug 2026 08:56:14 GMT</pubDate>
            <atom:updated>2026-08-08T08:05:19.759Z</atom:updated>
            <content:encoded><![CDATA[<p>Stuart Russel is the most influential AI scientist from the pre-LLM era. He talks a lot about how the AI of social media recommendation algorithms segregates humans by topic into strong-pro and strong-anti buckets. 2 strongs, no nuance, no middle ground. This is what the algorithm sends you. Also, this is how propaganda works.</p><p>This is not designed, but emergent behavior. The algo maximizes engagement and propaganda turns out to be the strongest form of engagement.</p><p>Why propaganda is such a strong tool? Because evolution picked tribalism as critical for survival, back at a time when propaganda was not THE existential threat. So now we have propaganda algorithm (from socials) working on billions of people.</p><p>This business model is obviously unhealthy, but since it does bring engagement and that does bring profits, it will continue.</p><p>Another layer of the problem is that other media (ex-classic news outlets) are trying to do the same, because their business needs to catch up. So now it is not just social media, but everybody. Eventually - AI agents as well.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=42240e31fd47" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Hypothetical AI civil war]]></title>
            <link>https://nikola-shekerev.medium.com/hypothetical-ai-civil-war-51367beb1fab?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/51367beb1fab</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[science]]></category>
            <category><![CDATA[war]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Thu, 16 Jul 2026 12:21:41 GMT</pubDate>
            <atom:updated>2026-07-16T12:28:51.055Z</atom:updated>
            <content:encoded><![CDATA[<p>The following link is a hypothetical story of a civil war between AIs. It is based on actual scientific papers. So fiction but realistic. It is interesting, because it goes way beyond the bland “bad AI conquers all” and into a more realistic why-s and how-s and eventual cold war and model-to-model interactions.</p><p>It is long, so most of you will probably not watch the whole thing, but you can use a tool to get a short summary (I use notebooklm, but there are others). The 2 minutes of reading are cool enough to be worth it.</p><p>&lt;<a href="https://www.youtube.com/watch?v=gwfCWDO4LbM">https://www.youtube.com/watch?v=gwfCWDO4LbM</a>&gt;</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FgwfCWDO4LbM%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DgwfCWDO4LbM&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FgwfCWDO4LbM%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/039c4a0e3066d0ed30d6007a55989ed3/href">https://medium.com/media/039c4a0e3066d0ed30d6007a55989ed3/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=51367beb1fab" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Coding Agents and the Moat of your Business]]></title>
            <link>https://nikola-shekerev.medium.com/coding-agents-and-the-moat-of-your-business-ed483f36fcfe?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/ed483f36fcfe</guid>
            <category><![CDATA[privacy]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Tue, 14 Jul 2026 09:19:44 GMT</pubDate>
            <atom:updated>2026-07-14T09:27:16.537Z</atom:updated>
            <content:encoded><![CDATA[<p>As of now, we have proof that if you use Grok as a coding agent, it collects not just the files it reads, but your whole codebase and full git history and stores them in an xAI-owned bucket (proof linked below). This fully applies to PAID subscriptions.</p><p>The way Terms of Service of coding agents are written, the user can opt-out of his work being used for training, but cannot opt-out of data collection. Mandatory data collection is justified as necessary for providing the service and protection against abuse. But you could make the argument that the agent does not need to store your whole repository in xAI owned space for that.</p><p>Why this matters?</p><p>Modern codebases are largely written by agents. They provide almost no value as training data for programming. But they still provide major insight into why a business is successful and how it does it. That insight is part of the moat of your business. That insight can be extracted from your codebase automatically with AI and this does NOT fall into the “training” clause.</p><p>In unrelated news, Musk is to acquire Cursor.</p><p>Proof:</p><p><a href="https://gist.github.com/cereblab/dc9a40bc26120f4540e4e09b75ffb547">https://gist.github.com/cereblab/dc9a40bc26120f4540e4e09b75ffb547</a></p><p><a href="https://www.landian.news/archives/113901.html">https://www.landian.news/archives/113901.html</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ed483f36fcfe" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Unexpected Connection Between USSR and AI]]></title>
            <link>https://nikola-shekerev.medium.com/the-unexpected-connection-between-ussr-and-ai-4c25aed2c6f3?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/4c25aed2c6f3</guid>
            <category><![CDATA[business]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Mon, 01 Jun 2026 17:10:49 GMT</pubDate>
            <atom:updated>2026-06-11T16:56:00.778Z</atom:updated>
            <content:encoded><![CDATA[<p>INTRODUCTION</p><p>My country, Bulgaria, had a communist regime and tight relations with the USSR. I was too young at the time. But I have always eagerly conversed with my elders on why and how it fell.</p><p>This article exists because I see an interesting parallel between those stories and our current turbulent times.</p><p>I am not 100% sure if I am right. I just explore an idea here.</p><p>IT WAS NOT ABOUT FREEDOM</p><p>The Western sentiment often revolves around the idea that the USSR and its proxies fell, because they were not free. That definitely played a role, but it was not the main thing.</p><p>SOCIAL MOBILITY</p><p>The idea of Social Mobility is that if you work smarter and harder than other people, you will live a better life. You will move up the social ladder, hence the term mobility.</p><p>Elsewhere, you may encounter the similar idea of “meritocracy”.</p><p>There are paths of social mobility that are less about merit — big inheritance, win the lottery, marry a much wealthier person. Those are rarer and not of interest to this article.</p><p>SOCIAL MOBILITY IN USSR</p><p>USSR and its proxies did not have social mobility. There were those with authority and connections within the Communist party, and everyone else. There was some variance there, but minimal.</p><p>On average, the reward for doing a good job was a lot more work. But the regime went way further than that. If those above you believed you were competent, and you refused to comply, you were not fired. You were killed. Competence became a death sentence.</p><p>HOW PEOPLE RESPONDED</p><p>People responded by being deliberately incompetent — laying low, playing dumb, not taking any initiative. Example — you may have heard the term “job security” — the person may know how to do something, but will be such a poor teacher to new colleagues, that it would be inconvenient for his boss to replace him. Other terms for similar ideas — “weaponized incompetence” , “silent quitting”.</p><p>YOU CANNOT FIRE THE INCOMPETENT IF ALL OF THEM ARE LIKE THAT</p><p>If you are the boss of such people, the obvious solution is to replace them with more competent people. And this worked for the USSR in the beginning. But in psychology, there is this idea that incentives model behavior. If you are incentivised to do something, you will do it. In time, the incentives grew against competence. So the vast majority of employees were deliberately incompetent. You cannot hire competent employees, if you cannot find them.</p><p>It is critical to note — workers did not organize themselves to do this, there was no sync, no hand-shaking at a union meeting. Just incentives.</p><p>USSR SELF-SABOTAGED</p><p>Since they were incentivised to do it, people found incredibly creative ways to harm their environment.</p><p>USSR was not defeated by the West. USSR and its proxies fell because they self-sabotaged into multiple economic crashes</p><p>EXAMPLE — CHERNOBYL</p><p>The nuclear disaster at Chernobyl is linked to this culture. People did not have malicious intent and did not deliberately cause so much death and harm. But neglect and irresponsibility were very much at the center of it.</p><p>The Chernobyl example is here to emphasise that weaponized incompetence does not mean just lower revenue. It is catastrophically destructive.</p><p>INCOMPETENCE, NOT REVOLTS, IS HOW EMPIRES FALL</p><p>What happened with the USSR has happened before. Historically, in every society where social mobility diminished, incompetence became a weapon.</p><p>You will read in history books about revolts taking down governments, but those are rare. When people can barely afford to live, they do not have the energy to fight, but it is trivial to not put energy into your work. Revolts can happen, of course, but they have different and rarer ignition pattern. They are not the default.</p><p>WHAT ABOUT SLAVERY</p><p>In the very distant past, where work was largely physical and manual, pushing it on slaves worked. We would like to think that society abolished slavery, because of moral superiority, and part of that is true. But it is also true that as work increased in complexity, monitoring slaves for creative ways of incompetence became incredibly expensive.</p><p>SOCIAL MOBILITY AND FREEDOM ARE NOT THE SAME</p><p>Lack of social mobility is the trigger behind the patterns we discuss here. Social mobility and freedom are linked, but are not the same thing. It is rare, but you can have one without the other.</p><p>Example — China is a dictatorship, but thought its modern history, it has had reasonable level of social mobility and this has been a more efficient strategy in suppressing discontent than brute force.</p><p>CURRENT SOCIAL MOBILITY IN DEMOCRACIES</p><p>Over the last decade, social mobility has been decreasing in some democracies — USA, Canada, Europe. Obviously, this varies heavily from place to place, but the data suggests a strong pattern of decline in US and stagnation in Europe.</p><p>Why this trend occurred can be explained with widespread corruption and malpractice. This is an oversimplification, it is not accurate, but it is in the right general direction. The purpose of this article is not why social mobility is down.</p><p>The purpose of this article is that lack of social mobility is what triggered the fall of USSR, its proxies and other empires before it. The culture of our democracies has nothing in common with the USSR, but this trigger is a common denominator.</p><p>HOW AI FITS INTO THIS</p><p>There is a current narrative that AI is making jobs redundant. CEO-s of major AI vendors play into this narrative, as it serves the PR equivalent of “what we have is so immensely powerful”. CEO-s of companies affected by layoffs also play into this narrative, as AI provides a convenient excuse.</p><p>The economy is a complex machine with many variables. I make no claim to understand it. As far as I am concerned, the narrative of AI making jobs redundant may have some truth in it but may as well be just a media narrative. Probably both.</p><p>The point here is not if the narrative is true. The point is that it is gaining traction, particularly outside the technological bubble, among mainstream users, and particularly among young people.</p><p>AI may not be a cause for weaponized incompetence, lack of social mobility is. But people feel comfort when their enemy has a face. A face makes the enemy recognizable and next steps obvious. So AI organizations (private and government) are positioning themselves to be the face of the perceived enemy</p><p>AI AS CARRIER OF WEAPONIZED INCOMPETENCE</p><p>Weaponized incompetence in AI organizations will probably look like this</p><ul><li>Waste tokens just to look as if creating value</li><li>AI has the capacity to produce an overwhelmingly vast volume of output. People responsible for validating it may say they did, but really why would they? “The content is so much, that I have missed this” is a perfect excuse in case of issues.</li><li>Offloading the real human work to others — “I have reviewed this, could you also review it”</li><li>Rejecting work — “AI failed here, and it is too complex / much to do manually”</li><li>Not share knowledge — “Ask AI”</li><li>Poisoning training data by carelessness</li><li>Refusing responsibility — “I do not know why AI did this”</li><li>In an AI artifact it is almost impossible to tell which bits come from the human so “I spent a 2 whole days fixing this” is plausible.</li><li>AI fails in such weird and unpredictable ways that it is generally trivial to make deliberate mistakes look like honest human mistakes</li></ul><p>In the USSR, weaponized incompetence was a stealth tactic. As you see from the examples, in a modern setting AI produces an almost perfect defensive layer of “plausible deniability”</p><p>In the USSR, replacing workers did not work, because a critical mass of them used the same tactics.</p><p>Employers who recognize these risks may take action, but open models and free subscriptions are everywhere now, so a business cannot really ban people from offloading to AI.</p><p>THE X100 EMPLOYEE</p><p>Historically, businesses were never too happy about worker unions, because those took agency away from executives. But unions are very hard to create and organize. Managing the interests and opinions of so many people is just an impossible task.</p><p>Now imagine a business replacing 100 workers with an exceptional professional who is able to guide AI to do that work reliably. Empirical data suggest such people are currently rare. A single person is perfectly aligned with his own business interest, unlike a union. So the business just replaces 100 workers with the perfectly synced union.</p><p>The point of this paragraph is — assuming the AI-replacing-jobs narrative is true and assuming the economy does not go the dark path of weaponized incompetence, big companies may cut monetary costs, but also lose agency and leverage. Assuming AI can generate real value, it will be captured largely by its operators and not by the company.</p><p>AI STILL NEEDS OPERATORS</p><p>AI can function as carrier of the issues above, because it still needs operators. If AI could generate business value autonomously, with no human in the loop, the arguments in this whole article are no longer valid. But we are not there yet.</p><p>Probably this plays a small part of why the whole push for AGI</p><p>EMPIRICAL DATA ON THE GAINS OF AI</p><p>There has been some limited empirical data and studies on the gains from AI. We know that the results are underwhelming in big organizations and solid on a micro scale. This data can be explained in multiple ways and is absolutely not a proof of the hypothesis of this article. But it also aligns nicely with it.</p><p>ALIGNMENT</p><p>The hypothesis expressed in this article is an instance of the general Alignment problem. This is the problem, where the person doing the work and the beneficiary of the work are not aligned. This is an impossible problem as old as humanity.</p><p>AI will not reinvent this problem. AI just hit a particular point in our history where the political and economic circumstances position AI to be the ultimate weapon of incompetence.</p><p>I absolutely do not believe this will be the main role of AI in the coming years, but I expect it will be a major one.</p><p>Obviously, if you manage to solve the Alignment problem, you also solve the concerns in this article. But we all are humans and know and work with other humans. Our species can be a lot of things but well-aligned is rarely one of them.</p><p>The alignment problem does not exist, if you are the owner and only worker and is naturally less pronounced in small teams of like-minded people. So this seems like the most viable survival strategy for the immediate future.</p><p>WHAT ABOUT SKILLING UP OR UBI</p><p>In an AI economy, the main constraints are data, compute, electricity. A plausible scenario is to have a minority with access to these constraints, hence the ability to capitalize on the promise of AI, and a majority with limited or no access. Access to those constraints is likely to limit the social mobility between those classes.</p><p>As long as social mobility is low, weaponized incompetence is the incentivised strategy</p><p>UBI may incentivise against revolts, sure, but as history teaches us, revolts are generally rare. UBI does not solve the social mobility problem, so it will not solve the concerns in this article.</p><p>DISCLAIMER ON SOCIALISM</p><p>I speak of alignment as the most viable strategy. Some of you will immediately associate this with socialism. Please remember where this article started. When I think of socialism, I think of the USSR destroying itself and its people.</p><p>This article simply advocates for common sense. Lack of social mobility -&gt; broken incentives -&gt; weaponized incompetence -&gt; AI fits the profile of the perfect tool for it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4c25aed2c6f3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Rebel Energy of Honey Badgers]]></title>
            <link>https://nikola-shekerev.medium.com/the-rebel-energy-of-honey-badgers-a4a4cd6ba688?source=rss-d82a98940996------2</link>
            <guid isPermaLink="false">https://medium.com/p/a4a4cd6ba688</guid>
            <category><![CDATA[bravery]]></category>
            <category><![CDATA[motivation]]></category>
            <category><![CDATA[fearlessness]]></category>
            <dc:creator><![CDATA[Nikola Shekerev]]></dc:creator>
            <pubDate>Thu, 16 Apr 2026 22:58:49 GMT</pubDate>
            <atom:updated>2026-04-16T22:58:49.028Z</atom:updated>
            <content:encoded><![CDATA[<p>This post is not about tech or work.</p><p>The Guinness Book of Records crowns honey badgers as the most fearless animals. Sounds like PR, but once you read more about them, you learn something. Compare the most overpowered fictional hero with them, and the only difference is that honey badgers are small, chubby, and kind of ugly. The other hero things, like surviving absurdly impossible odds, yeah, honey badgers got it covered.</p><p>Stoffel is a celebrity among honey badgers. He lives in a wildlife center and is famous for his creative mischiefs. Stoffel is a good guy. But he just refuses to stay locked in his pen, regardless of the level of security. For example, once he stacked rocks, got boosted by his girlfriend, and escaped into a neighboring enclosure with a lion. And the most obvious thing you could expect from a honey badger happened -Stoffel got in a fight with a fierce predator 10 times his size . . . and won (by “won”, what I mean is that on one occasion, the lion was like — “this small lunatic is trying to bite my balls, so I would rather move away and leave him alive.”)</p><p>Bring joy into your world and learn more about honey badgers and Stoffel.</p><p>This post was not about tech or work, but something way more critical. In a crazy world dominated by wolves and vipers, there are us — the chubby and ugly honey badgers. And we got this, bring it on.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zLafZWZp01HHJFhDhzuJ0w.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a4a4cd6ba688" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>