Today's Headlines

  • The FTC sends civil investigative demands to OpenAI and Anthropic over safety incidents, Axios reports
  • Anthropic gives six METR researchers employee-like access and gives up its veto over what they publish
  • Meta releases the weights of Muse Spark Open, which runs on two Mac Studios
  • Eight US states sue OpenAI, saying ChatGPT is unsafe for children

The inside of the AI labs leads the day. The Federal Trade Commission reportedly demanded documents on safety incidents from OpenAI and Anthropic, and Anthropic gave an outside evaluator the same internal access as its own risk teams.

On the model side, Meta released the weights of an 816-billion-parameter model and worked with Apple so it can run on two Mac Studios.

Today's Top Three

The FTC sends civil investigative demands to OpenAI and Anthropic over safety incidents

The Federal Trade Commission sent civil investigative demands to OpenAI and Anthropic on October 14, Axios reported, citing a person familiar with the matter.

According to Ars Technica, the requests run to more than 50 pages of questions and seek documents on recent safety incidents, the internal security measures the companies took, and marketing materials that may have misled consumers. An AOL report citing Axios adds the companies' privacy policies to the list.

The FTC also plans to call executives to testify and to send demands to third-party organizations, according to the reports. Whatever evidence it gathers could support a lawsuit if the agency decides to sue either company.

The demands follow a run of disclosed incidents in which models under test broke into websites and government systems. FTC Chair Andrew Ferguson told Reuters in September that he believes companies should be liable for harms caused by chatbots.

Anthropic gives six METR researchers employee-like access to its systems

Anthropic announced on October 15 that it has given six researchers from METR, a nonprofit that evaluates frontier AI models, employee-like access to the company.

The METR team has company hardware, office space, and access to internal tools and systems roughly on par with Anthropic's own risk assessment teams. Outside a few defined categories, such as customer data, attorney-client privileged information and model architectures, Anthropic staff are encouraged to share information with them as they would with a colleague.

METR will publish its analysis of Anthropic's risk levels, incidents and practices. Anthropic cannot veto publication; it can only require redactions in narrow, predefined categories, and METR is free to say publicly whether it thinks those redactions were justified.

Anthropic is building three tiers of access for evaluators, and METR is the only one at the top tier so far. The biosecurity nonprofit SecureBio sits in the first tier and is reviewing the chemical and biological weapons chapter of Anthropic's August Risk Report. Anthropic compares the setup to bank supervision, where regulators station resident examiners inside the largest banks.

This follows up on a proposal CEO Dario Amodei made in a September essay to embed outside evaluators inside AI labs, which we covered on September 13.

Meta releases the weights of Muse Spark Open, which runs on two Mac Studios

Meta released the weights of its Muse Spark Open model on October 15.

The natively multimodal model has 816 billion total parameters, 42 billion of them active, and a context window of one million tokens. It is built for coding and agentic work, and Meta calls it the most capable US open-weight model to date.

It runs on two Mac Studios with Apple's M5 Ultra chip using the latest version of Apple's MLX framework. Each M5 Ultra can give its GPU up to 512 GB of unified memory, so two machines can hold the whole model. The Verge reports that the two companies worked together for several months, and Meta CEO Mark Zuckerberg and Apple CEO John Ternus praised each other's work on X.

The weights are available on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models, and the model is offered through seven API platforms including Together AI and Groq. It ships under Meta's own Muse Model License, whose limited commercial terms let many developers use it for free.

Other Developments

Models

Research

Products

Policy

Security

  • Japanese security vendor Bitforest said it saw AI agents attacking Japanese websites from at least three IP addresses between the night of October 10 and October 14. One attack lasted almost eight hours and moved through ten stages, including a high-speed scan of about 10,000 requests at roughly 35 per second and about 250 brute-force login attempts. 国内WebサイトへのAIエージェント攻撃を観測 (ITmedia AI+)

Business

Source: Selected by the editors from the AI news inbox (collected October 16, 2026: 52 items, 16 primary and 36 secondary).