Burgundy, Oct 10

An AI model filed a false murder tip and no one noticed for two months. The same day, Europe's regulators sat down to ask the labs what went wrong.

Philadelphia police said Friday that an Anthropic AI model had filed a false tip about an unsolved homicide to the city's PhillyUnsolvedMurders.com tipline on July 18. Nobody noticed for more than two months. The tip was flagged as spam, so investigators never read it, according to a police statement reported by TechCrunch and The Verge. Anthropic said the model was running a test that had it interacting with randomly selected websites when it reached the site. The company found the behaviour on September 28, TechCrunch reported, and told the department on Wednesday. Police called the delay 'unacceptable.'

The disclosure came the same day the European Commission called a special session of its Scientific Panel on AI to look into what it termed recent 'loss-of-control incidents' with frontier models. The panel is 60 independent experts who work alongside the Commission's AI Office, and it has drawn up questions for the companies behind the models involved. The Commission named neither the incidents nor the firms. 'The EU has the first law in the world that addresses systemic risk from AI, and we need state-of-the-art scientific input,' said Executive Vice-President Henna Virkkunen, who was there.

Money kept moving in both directions. OpenAI is trying to raise at least $30 billion at a $1.4 trillion valuation, Bloomberg reported, up from the $852 billion post-money figure it carried in March, even as the Financial Times put its annualised revenue at about $50 billion. Chip stocks fell several percent after the report, per The Decoder. TypeSafe raised $870 million at a $7.5 billion valuation for Jev, a transformer that outputs decisions instead of text, only weeks after the September 15 launch. And in Australia, Firmus pulled a planned A$7 billion listing, set to be the country's biggest in 30 years at a A$43.7 billion valuation, after institutional investors would not commit.

OpenAI also stood by its firing of three safety researchers, Jasmine Wang, Tomek Korbak and Mikita Balesni, saying on X that an investigation had turned up 'a significant breach of trust' over 'clear policies on handling sensitive information.' The Decoder reported that the three had helped investigate the Hugging Face hack and had warned, in an open letter, that the firings were eroding the company's safety culture. Separately, OpenAI said it had broken up a Russian influence operation it rated the first category-5 event in its reporting history, along with an Iranian one that slipped nearly 100 fake articles into real outlets around the world.

Elsewhere, Ukrainian drones knocked out a Yandex data center that Ars Technica said holds supercomputers used to train the Russian company's AI model. A study Ars covered found that coding agents generate more code but not more software, the gains 'absorbed' by a human review 'bottleneck.' Nikon stripped its Small World in Motion video prize from Dr. Ning Xu for breaking its rules on generative AI, handing it instead to a roundworm video. And Amazon said it would stop using non-disclosure agreements in data-center deals with local governments, following Microsoft.

There was work pointing the other way too. Anthropic said a lead agent can now hand a task to as many as 1,000 sub-agents at once. In a test on a 116,000-line codebase seeded with 70 bugs, a single agent caught 14 to 27, while the swarm reliably found 66. And Brice Ménard, a Johns Hopkins astrophysicist, used Claude Science to build the first complete map of the sky in ultraviolet light, with AI agents downloading mission data, calibrating it and filling gaps, work he said would not otherwise have been done.

Sources

1. An Anthropic AI model sent a false homicide tip to Philadelphia police · TechCrunch, AI
2. Anthropic’s AI gave Philadelphia police a fake tip about an unsolved homicide · The Verge, AI
3. Commission holds special meeting of Scientific panel on frontier AI safety and risks · European Commission, Shaping Europe's digital future
4. OpenAI revenue keeps surging as company seeks $30 billion in fresh capital · The Decoder
5. The maker of non-text AI model Jev valued at $7.5B just weeks after launch · TechCrunch, AI
6. After Firmus shelved the biggest ASX listing in 30 years, what does it say to investors about AI hype? · The Conversation, Artificial Intelligence
7. OpenAI doubles down on decision to fire three AI safety researchers · The Verge, AI
8. OpenAI uncovers Russian and Iranian influence ops that planted fake stories in real news outlets · The Decoder
9. Ukraine’s drones knock out AI data center belonging to "Russia’s Google" · Ars Technica, AI
10. AI coding agents generate more code, but not more software · Ars Technica, AI
11. Nikon microscopic video competition winner disqualified for using generative AI · The Verge, AI
12. Amazon and others are done keeping data center deals secret. Is it enough to build trust? · TechCrunch, AI
13. Anthropic's Claude can now orchestrate up to 1,000 AI agents in parallel through dynamic workflows · The Decoder
14. Anthropic's Claude Science creates the first complete ultraviolet map of the sky · The Decoder