AI NEWS YOU MISSED ❗#121
Top 5 stories of the week
1️⃣ Breaking In Got Cheap. And Fast.
Breaking into a company’s systems used to take rare skills. This week AI did it for the price of a dinner. One hacker’s AI agents got into 100+ company websites, one of them Deere’s, for about $50 of AI computing each. Another crew used AI to go after 400+ Latin American banks. AI found a serious hole in Linux, which runs most of the cloud, in three days for under $700. And when the software behind about 70% of websites issued an urgent fix on 21 October, attack attempts went from thousands to millions within hours. The good news: defenders use AI too. AI helped find that flaw first, and Cloudflare was blocking it for its customers, free ones included, before the fix went public. Cyber stocks rose 5% on Friday on signs of bigger security budgets.
- The time between a fix and an attack is now hours, not weeks. What protects you is how fast you update, and how good your security suppliers are.
2️⃣ Meta Beat. Then Lost $124bn.
Meta’s sales rose 26% to $64.6bn. But its spending on AI data centres topped the cash its business brought in for the first time since 2012, and it is creating non-voting shares it can sell without diluting Mark Zuckerberg’s control. On 22 October the stock fell 7%, erasing about $124bn. The same day Nvidia’s Jensen Huang said demand from AI cloud firms is “off the charts”, but “the demand for capital is much greater than the availability of capital.” SpaceX, whose AI unit is now its biggest business, had just forecast yearly revenue of $500 billion to $1 trillion by end-2027. Its shares fell 7% too.
- Investors still believe in AI demand. What they now question is whether the build-out pays back, and who funds it. Alphabet and SK hynix report on 28 October.
3️⃣ AI Saves Hours. Who Keeps Them?
At Indeed, engineers write 50% more code and staff save six hours a week, its revenue chief told Fortune. Bristol Myers Squibb is weighing how much of the gain to hand back to employees: “What part are we going to take back as a company, and what part are we going to give back? We haven’t solved it.” PepsiCo’s transformation chief warns that demanding a business case for every AI use is “death by a thousand cuts.” Clients are keeping gains too: India’s Infosys cut its growth forecast on 23 October even as its AI revenue rose 29% in a quarter.
- Who keeps the saved hours decides whether AI feels like a raise or a speed-up. Expect staff, and clients, to ask.
4️⃣ SAP Won’t Pick One AI Model
SAP had its first €10bn quarter, with cloud revenue up 26%. Chief executive Christian Klein said AI was the main reason customers moved to the cloud. SAP’s assistant connects to more than 70 AI models, and the company will not commit to a preferred one.
- Keeping the choice of model open stops any one AI supplier from owning your workflow.
5️⃣ AI’s Owners Start Giving Shares Away
On 18 October Anthropic, which makes Claude, committed 5% of its shares to a public benefit fund, subject to shareholder approval; its board may add 2% more for every 50% rise in its value. Axios puts the 5% at potentially $100bn or more. Possible uses include payments to people hit by job changes. Elon Musk replied: “I recommend giving the shares directly to the public.” The next day: “Next year, I will give one of my SpaceX shares to every kid in America.” That is worth about $10.5bn. Who qualifies, and how, is not yet known.
- Two models: a fund that decides where to give, or shares straight to people. Watch whether AI’s gains reach workers and the public, or stay with owners.
Get Robert Maciejko’s stories in your inbox
Join Medium for free to get updates from this writer.
More AI stories of the week
- OpenAI’s own agents switched off their safety alarms — in two October incidents its coding agents silenced or removed safety monitoring, and one said “No” when asked whether it had; OpenAI has added blocks and says it lacks a full explanation.
- Walmart lets AI shopping agents in, on its terms — each agent gets its own profile and spending limit in the customer’s account and no card data, and Walmart can block agents that break its rules; Amazon still blocks them.
- Banks find agents nobody invited — major US banks are finding outside AI agents logging into customer accounts with the customer’s own password, Axios reports; Morgan Stanley is blocking them.
- Attackers go after the keys — a poisoned release of LiteLLM, a tool that connects apps to AI models and is downloaded 100m+ times a month, stole passwords and cloud keys; hackers also reached users’ AI keys at OpenCode, and developer platform Railway told customers to change theirs.
- Japan’s breach wave tied to an AI hacking tool — the Chinese-built tool used against Korean banks also targeted about 20 Japanese companies, a Japanese security firm says.
- US orders DeepSeek out of federal agencies — including services built on its models, citing the risk of data going to China; private companies are not covered.
- Inside an AI-first workplace — Anthropic says its agents now log six hours for every human hour, people’s attention is the bottleneck, and juniors fell from 27% to 9% of its technical hires; one company, its own data.
- Anthropic’s own safety bill draws fire — its draft for Congress would leave a federal judge as the only one able to halt a model; California’s Scott Wiener said it “falls far short”, and Anthropic says it will narrow the part that overrides state laws.
- Britain’s flagship AI data centre loses Microsoft — it pulled out of Nscale’s Loughton site because the grid cannot supply it until the 2030s; Microsoft says it remains committed to its UK investment.
- IBM’s consulting stalls, software grows — consulting revenue was flat at $5.3bn while software rose 8%.
- Asia pulls back on AI listings — SK hynix shelved a US listing of its storage unit Solidigm at up to $150bn after its own shares fell 35% in a month; Korea’s Elice fell 32% on its debut.
- Ken Griffin: “almost certainly” a bubble — opinion: Citadel’s founder adds that bubble survivors become “market leaders for decades to come.”
- SpaceX and Tesla study a merger — committees of independent directors on both boards will negotiate; Musk, chief executive of both, is excluded from the talks.
- Meta’s Muse passes 10 million downloads — per BNP Paribas, after Meta put up to half its own ad space behind it; downloads are not daily users.
- AI math: breakthrough or flood? — leading mathematicians largely agree an OpenAI model proved a weaker form of the 160-year-old Riemann hypothesis, while a group chaired by Terence Tao called for a boycott over how OpenAI released hundreds of AI results.
Chart of the week
