Anthropic pledges 5% of shares as FTC demands its records
Anthropic will commit 5% of its shares to a public benefit fund, subject to stockholder approval, the same week the FTC demanded its safety records.
Two kinds of governance landed in the same week
Anthropic said on October 18 that it intends to commit 5% of its shares to a wholly owned LLC called the Anthropic Public Benefit Fund, subject to stockholder approval, with its board able to add 2% more for every 50% rise in valuation above a baseline the company has not disclosed, and the equity barred from lobbying or political giving (CNBC).
Three days earlier the FTC sent civil investigative demands running more than 50 pages each to OpenAI and Anthropic, seeking records on security incidents, safeguard practices, marketing materials and privacy policies, and said it plans to compel executives at both companies and at the research group METR to testify (Axios).
Both are governance aimed at the same two companies in the same week. Only one of them carries a date.
The voluntary instrument is denominated in a moving number
The fund is sized as a share of equity, not as a sum of money. Anthropic's last marked valuation is $965B, set in the $65B Series H in May 2026 (Anthropic), so what 5% is worth depends entirely on when it is counted. The escalator makes that dependence explicit: more equity arrives only as the valuation climbs 50% above a baseline that has not been published, which means the fund's eventual size cannot be read off the record as it currently stands.
Two conditions sit in front of it as well. The commitment is an intention subject to stockholder approval, and Anthropic filed confidentially for an IPO in June (TechCrunch), the same month OpenAI filed with the SEC (CNBC). Which register of stockholders votes on a 5% transfer, and when, is the part that decides whether this is a structure or a statement.
The compulsory kind names dates and penalties
Three measures in the same week put numbers on the calendar instead.
- FTC: demands of more than 50 pages each, for documents that already exist, plus planned compelled testimony from executives at both labs and at METR (Axios).
- Taiwan: AI servers and advanced chips move onto the controlled export list, making unlicensed shipments to China a criminal offence carrying up to five years in prison, a step the economy minister tied to commitments in the reciprocal trade agreement with the US (Bloomberg).
- Microsoft: the UK Competition and Markets Authority proposed strategic market status for its business software, after provisionally finding high switching costs and bundling across operating systems, productivity, security, server operating systems and databases used by around 20 million monthly UK users, with consultation open until November 13 and a final decision due by February 2027 (Computer Weekly).
The shapes differ. One asks for files, one threatens prison, one opens a consultation and fixes a decision roughly four months out. What all three share is that none of them depends on the subject agreeing.
What to check next
- Whether Anthropic publishes the valuation baseline its 2% escalator keys off, and whether stockholders vote before or after a listing. Both are unresolved in the only account of the fund on the record (CNBC).
- Whether METR testifies. A regulator compelling an evaluator as well as the developers is a wider reading of who answers for a model's safeguards than anything the labs have volunteered.
- The CMA's November 13 consultation close, then its February 2027 decision. That is the first date on this list that would force a change in a product rather than in paperwork.
The equity commitment may well prove the largest of the four in money terms. It is also the only one of the four that its subject can revise, defer, or quietly decline to finish without needing anyone else's permission.
Built from the digest of 2026-10-19.