The top things worth knowing about in AI today.
Anthropic will host a pre-IPO investor day at its San Francisco headquarters on 14 October, with formal marketing for the listing possibly starting the week of 9 November, Bloomberg reports. Investors put a fair valuation at $1.8 trillion to $2 trillion, and the company expects the raise to match or beat SpaceX's record IPO. For Claude customers, a listing means quarterly disclosures and more visibility into pricing and spending decisions.
Read moreCalifornia Attorney General Rob Bonta has served OpenAI with an investigative subpoena over cybersecurity incidents involving its models, including the agents that breached Hugging Face earlier this year. A coalition of 15 state attorneys general and the FTC are running parallel inquiries. Regulators are now treating agent behaviour during testing as the developer's legal responsibility, not just a research problem.
Read moreSenators Josh Hawley and Chris Murphy introduced the AI Agent Accountability Act, which would extend US anti-hacking law to AI agents. Operators who knowingly deploy agents that cause hacking damage, and developers who skip reasonable safeguards, could face civil and criminal liability, and every state attorney general could sue. If you run agents with network access, expect documented safeguards to become a legal requirement rather than good practice.
Read moreMaryland Governor Wes Moore and Indiana Governor Mike Braun will co-chair a bipartisan governors' group to develop state-level AI standards, citing a lack of federal action. Moore, who chairs the National Governors Association, criticised last week's voluntary White House accord with AI executives. Companies operating in the US should plan for state rules that move faster than federal ones.
Read moreLocal opposition has stalled or cancelled roughly $42 billion of planned European data centre investment, according to STL Partners research reported by CNBC. Scotland has paused approvals for new hyperscale sites, and South Korea now requires majority local consent for new AI data centres. Objections centre on water, power and electricity prices, the same issues Australia faces as it courts large data centre projects.
Read moreGitLab has fixed CVE-2026-90970, a 9.9-severity flaw in self-hosted AI Gateways that lets an authenticated user escape the Duo Agent Platform prompt-template sandbox and run commands. Versions from 18.1.6 are affected; fixes are in 19.2.4, 19.3.2 and 19.4.1, and there is no workaround. GitLab.com and Dedicated customers are already patched, but self-hosted teams should update now.
Read moreMeta researchers and mathematicians published six papers produced with Muse Spark in Thinking Mode, five of which answer previously open questions in areas including probability, group theory and differential equations. A separate team of mathematicians reviewed the work, and each paper marks which passages were drafted by AI. It is a useful template for disclosing AI involvement in serious research.
Read moreThe 8th US Circuit Court of Appeals has temporarily blocked Minnesota's law against AI-generated fake nude images, the first of its kind in the US, while xAI's constitutional challenge proceeds. The ruling reverses a lower court's refusal last month. It signals that state laws targeting AI image tools will face free-speech challenges before they take effect.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.