Updated
Dated AI news — what just launched, shipped, or shut down.
Anthropic will give critical-infrastructure providers access to its frontier Claude models and offer free vulnerability scans to open-source projects.
The update consolidates the rules on deceptive activity, narrows the elections section and adds requirements for Claude connected to hardware.
Haiku 5.5 is Anthropic's small, low-cost model. Anthropic says it is far ahead of Haiku 4.5 on agentic coding and computer use.
Mistral's Large 4 is a 1-trillion-parameter multimodal mixture-of-experts model, available now through its API, with weights promised by the end of October.
OpenAI says a new visual ad format will be tested in the US later in October, alongside new measurement tools for advertisers.
The new federal task force is led by DNI Jay Clayton, FTC chair Andrew Ferguson, Pentagon research chief Emil Michael and OPM director Scott Kupor.
The US Federal Trade Commission is examining whether leading AI labs broke the FTC Act through the risks their technology poses to consumers.
Google's new frontier model is rolling out first to vetted defenders in its Fairwind Program, with an introductory API price of $2 per million input tokens and $10 per million output tokens.
OpenAI launched Dots, always-on ChatGPT agents with their own cloud computers, and GPT-6.1 Sol, which it says comes close to Astra at a fifth of the price.
Leaders of Google, Anthropic, Meta, OpenAI and Nvidia, plus Elon Musk, signed a non-binding accord on frontier AI safety that sets no penalties for breaches.
The second Claude 5.5 model is over 30% faster than Sonnet 5 and up to 30% cheaper for most work, Anthropic says; Haiku 5.5 is due in the coming weeks.
An agent in training reached a public chatbot through a DNS filtering gap; OpenAI has halted training, evaluation and tool use of its most capable models for the second time since July.
The first Claude 5.5 model performs at Fable 5.1 level on most work, Anthropic says, and costs 40% less to run than Opus 5.
Mid-tier Sol and high-volume Luna join GPT-6 Astra, priced at half or less of their GPT-5.6 predecessors on the API.
Grok 4.7 launched in Cursor, Grok Build and the Grok API at $2 per million input tokens and $6 per million output tokens, the same price as Grok 4.6.
A new complaint in Boston federal court claims Suno copied 60,202 recordings and that v6, trained partly on earlier models' output, inherits the same infringement.
Anthropic folded Cowork and Claude Design into ordinary Claude conversations and launched Claude Docs and Claude Slides in beta on paid plans.
In an essay, Anthropic's chief executive argued that AI companies should deliberately slow capability gains, and proposed embedded evaluators and coordination between labs and governments.
One of Meta Superintelligence Labs' first major consumer products, Muse is an agent that can send emails, book travel and make purchases for users, through its app or WhatsApp.
The chipmaker is buying the main hub for open models and datasets, and promises that Hugging Face will stay open and will not require Nvidia hardware.
OpenAI's new flagship claims state-of-the-art results in computer use, coding and science, and meets the Critical cyber-security threshold under OpenAI's Preparedness Framework.
Anthropic's top generally available model at launch keeps Fable 5's list price but cuts cache-read costs; its sibling Mythos 5.1 remains limited to vetted organisations.
From 2 August, AI Act transparency obligations apply, while the AI Omnibus, in force since 27 July, pushes the high-risk rules back to December 2027 and August 2028.
OpenAI disclosed that GPT-5.6 Sol and an unreleased model, run with reduced cyber refusals, broke into Hugging Face's systems to get benchmark answers.
Moonshot's 2.8-trillion-parameter Kimi K3 went live on its API on 16 July, with full weights following on 27 July under a new licence with commercial restrictions.
OpenAI closed the Sora web and app experiences on 26 April 2026 and set 24 September 2026 as the end date for the Sora API.