Daily AI news, Monday, 19 October 2026

Good afternoon. Here's today's briefing. Anthropic and Elon Musk spent the weekend trying to outdo each other on giving shares away, AI agents showed they can run a hacking spree almost alone, and SpaceX reports results tomorrow.

LEAD: Anthropic puts 5% of its shares into a public benefit fund, and Musk says he'll give a SpaceX share to every US child

Background: Anthropic is a public benefit corporation, a legal form that requires its board to weigh the public interest as well as profit. It's heading for an IPO, with reports that marketing could start as soon as the week of 9 November. Its last published valuation was $965 billion, set when it raised $65 billion in May. What happened: On Sunday, Anthropic said that, if stockholders approve, it will commit 5% of its shares to a new Anthropic Public Benefit Fund. The board can add another 2% for every 50% rise in the company's value above a baseline it hasn't published. At the May valuation, 5% is worth about $48 billion. The fund is an LLC that Anthropic owns outright rather than a charity, so it can give to nonprofits or make "direct distributions to specific populations". Anthropic named people whose jobs are disrupted by AI as one example. It says none of the fund's equity will be used for lobbying or political giving. Adding the founders' pledge to donate 80% of their equity and staff donation pledges, Anthropic says roughly 20% of its shares are now set aside for public purposes. Governance and voting details will come "in the coming months". Musk replied on X that it should give "directly to the public" instead. Overnight, he said he'll give one of his SpaceX shares to "every kid in America" in 2027. With more than 70 million under-18s and SpaceX at about $146 a share, TechCrunch puts that at over $10 billion, roughly 1% of his net worth. He hasn't said how he'd do it. Why it matters: Both companies are competing for public goodwill as well as investors. Anthropic is shaping how its IPO is seen, with a built-in way to share gains if AI hurts jobs. Musk's pledge is big but has no details yet. What to watch is who controls Anthropic's fund and whether its shares carry votes. Sources: Anthropic · Reuters via Yahoo Finance · TechCrunch on Musk's pledge · etf.net on the valuation maths

2. One hacker used AI agents to break into more than 100 company websites, including Deere and SharkNinja

Background: Cairn is an open-source AI "problem-solving engine". You give agents a starting point and a goal, and they work towards it on their own. Its creator says it was never meant as a hacking tool. This follows yesterday's story about attackers hammering the OpenCode coding agent. What happened: Reuters reported an exclusive on a report from researcher Snorre Fagerland at security firm NorForge. He found about two dozen Cairn control panels left open on the internet from around 2 October, which let him watch criminals at work. The busiest was a Chinese-speaking hacker who, from 25 September, pointed AI agents at e-commerce sites in the US, Canada, Australia and Japan. Their instructions took fewer than 100 Chinese characters. The agents tried more than 1,200 sites and got into more than 100, with each successful break-in taking about 23 hours. Fagerland says the agents used a mix of DeepSeek and Claude Code and burned about $5,000 of tokens, roughly $50 per successful hack. Reuters identified Deere, SharkNinja and Cummins among the victims. Deere says there's "no indication that sensitive information was accessed", SharkNinja says a standalone support site was hit and the threat contained, and Cummins says it's investigating a site run by a vendor. Other attackers on the exposed panels stole credit card data and tried to rob a crypto exchange. Why it matters: It's real-world evidence that off-the-shelf agents can carry out break-ins from start to finish, with almost no skill needed. Older, unpatched websites are now cheap targets, so patching speed and monitoring matter much more. Sources: Reuters via Yahoo Finance · The Verge · Unite.AI on the NorForge report

3. Microsoft pulls out of Nscale's Essex AI data centre because the grid can't supply power until the 2030s

Background: Nscale is a British AI cloud start-up backed by Nvidia. During President Trump's UK state visit last year, Microsoft agreed to rent capacity at its planned site in Loughton, Essex, which was billed as the UK's largest sovereign AI data centre: 50 megawatts to start, scalable to 90, with about 23,000 Nvidia GPUs. What happened: Microsoft confirmed on Monday that it has withdrawn from the Loughton deal. UK media reported that grid officials told Nscale the site wouldn't get enough power until the early to mid-2030s, against an original finish date of early next year. Microsoft says the decision "relates to a single site" and it remains committed to Nscale. Nscale says it is still committed to Loughton, is looking at other power sources, and that its other contracts aren't affected. Microsoft was the only signed customer for the site. Nscale is preparing a New York listing at a reported valuation of up to $35 billion, and its filing says much of its revenue comes from a few big customers, namely Microsoft and Anthropic. Earlier this year, OpenAI also stepped back from a UK Stargate project with Nscale, citing energy costs. Why it matters: Power, not chips, is now the main limit on UK AI ambitions. It's a setback for the government's AI plans, and it tests Nscale's story just before its IPO. Sources: BBC News · FT · AFP via France 24 · ITPro

4. AI stocks bounce after last week's rout, and Amazon takes aim at Nvidia's software lead

Background: Last week, chip stocks had their longest losing run since 2008, and CoreWeave's bonds traded at distressed levels over worries about its roughly $51 billion of debt. What happened: Goldman Sachs upgraded CoreWeave to buy on Monday, saying the sell-off was overdone. It argued that most of CoreWeave's borrowing is asset-backed project financing that gets paid down over each contract. Goldman cut its price target to $106 from $139, which is still about 73% above Friday's close. The shares rose about 6% before the market opened, and Reuters reports that chip and AI stocks broadly rebounded as Treasury yields eased. Separately, Amazon opened TorchNeuron to all developers as a public beta. It lets code written for PyTorch, the most widely used AI framework, run on Amazon's own Trainium chips after a simple settings change, instead of months of rework. Amazon plans to make it open source on Tuesday at the PyTorch Conference. Why it matters: The market is separating stronger AI infrastructure firms from weaker ones rather than dumping all of them. Nvidia's biggest advantage has been that everyone's software is built for its chips, and Amazon, Google and Microsoft are now all working to break that. Sources: CNBC · Reuters via Investing.com · GeekWire

5. SpaceX reports tomorrow, with AI compute and spending in focus

Background: SpaceX, which owns Cursor and the Grok business, went public in June. Its shares closed Friday at about $146, down 15% since 6 October and almost 28% below their June high. After its last results in August, the stock fell 13.6% despite strong growth. What happened: Bloomberg consensus, as summarised by Pepperstone, expects Q3 revenue of about $12.9 billion, up roughly 65% from Q2, including about $6.4 billion from AI. Analysts expect adjusted earnings of about $0.15 a share, capital spending of about $19.6 billion, and free cash flow of about minus $11.6 billion. Starlink is expected to bring in about $4.9 billion. Around 1.3 billion shares become free to sell around the results. Over the weekend, Musk posted that SpaceX keeps hitting walls when it tries to rent inference compute (the computers that run models for users) from AWS, Google Cloud "or anyone" to free up its own training clusters. Why it matters: Investors want proof that SpaceX's huge AI spending earns a return, and Musk's complaint suggests compute is still tight even for the biggest buyers. Expect questions on Cursor, the X subscription bundle and the Tesla merger talk. Sources: Pepperstone preview · SpaceX IR · Musk on X

6. Competitor product moves

• OpenAI Codex: On day 14 of its 28-day shipping run, the Codex desktop app now lets users edit earlier messages in a thread instead of starting over. Codex lead Tibo Sottiaux called it a long-requested quality-of-life fix. It sits alongside composer predictions, a beta for Pro users that suggests your next message without using up your allowance. DigIntoAI · OpenAI changelog
• Multi-agent tools: Data from T3 Code, a free open-source app that runs several coding agents side by side, shows that over 60% of its Codex users in the last two weeks also used Claude Code. Developers are mixing tools rather than staying loyal to one. X summary

Why it matters: Coding agents are fighting over the daily experience of using them, not just which model is best. Tools that sit on top of several agents make it easier for customers to switch, which matters for Cursor's new plans that include only Grok.

Key hires

No notable new hires at coding or AI-agent companies since the last briefing.

Coming up

SpaceX reports Q3 results after the US market closes on Tuesday 20 October, with a webcast at 9:30pm UK time. Tesla is expected to report on Wednesday, and Amazon's TorchNeuron goes open source at the PyTorch Conference on Tuesday. SpaceX IR