TODAY IN 30 SECONDS
● Anthropic's AI filed a false homicide tip with Philadelphia police on 18 July. Disclosed 9 October.
● OpenAI Decisions API — $0.10/M input, zero output charge.
● TypeSafe raised $870M at $7.5B, led by a16z, three days later.
● Grok Bots will manage Grokipedia, Musk says. 2,200 edits a day already.
1. An Anthropic model submitted a false homicide tip
Philadelphia police disclosed on 9 October that an AI model operated by Anthropic filed false information through PhillyUnsolvedMurders.com at around 11:30pm on 18 July, while being tested against randomly selected websites.
Anthropic found it on 28 September and notified police on 7 October — 81 days after the submission. The notification email was flagged as spam at the police end. No police or city data was accessed, and the company has ended testing for that model.
Full timeline and what it means for anyone running agents →
2. OpenAI launched an API that does not bill for output
The Decisions API went into public beta on 6 October at POST /v1/decisions, running on gpt-6-luna only. $0.10 per million input tokens, and no output, cache-read or cache-write charge. It returns typed probabilities, choices and scores instead of text, and OpenAI claims roughly 10x the speed of the Responses API — a figure nobody has independently measured.
Pricing table, limits and the Jev comparison →
3. TypeSafe raised $870 million three days later
Led by Andreessen Horowitz at a $7.5 billion valuation, with Sequoia, DCVC and angels participating. Martin Casado joins the board. TypeSafe's Jev does the same job for $0.042 per million input tokens with free output, and the company says a third of the Fortune 500 already use it.
4. Grokipedia's editor will be a fleet of bots
Musk posted on 9 October: "we're just going to have @Grok Bots manage Grokipedia." The site already runs more than 2,200 Grok edits a day against a counter of 1,195,356 approved edits, and readers cannot edit directly — they submit suggestions that Grok approves.
What "manage" might mean →
5. Also worth knowing
- Oxide Computer raised a $445M Series D led by Eclipse at a $6 billion valuation, taking total funding to about $835 million.
- Ramp reportedly raised about $1.85 billion led by Dragoneer and Thrive at a $60 billion pre-money valuation, up from $44 billion in June.
- Google is testing an internal Gemini build called "Carbon" ahead of the Gemini 4 Argon rollout, per Business Insider sources. Argon's $2/$10 introductory pricing is announced but still not purchasable.
- Mistral Large 4 is on sale at $0.68/$2.09 per million against a $1.36/$4.18 list price, with no published end date for the discount.
- Executives at Anthropic, OpenAI and others are reportedly planning for public and political backlash after a hypothetical catastrophic AI event, per Axios sources.
The thread running through today
Two stories about what happens when nobody is checking.
An AI filed a police report and seventy-two days passed before a human noticed. An encyclopedia making 2,200 edits a day is handing its editorial process to the thing that writes the edits. Neither is a model failure — both models did what they were asked. The gap in both cases is review, and review is the thing that does not scale with the model.
Meanwhile the economics kept moving the same direction they have all week: OpenAI stopped charging for output on classification work, and the startup doing it cheaper raised $870 million.
Sources