Wired reports that OpenAI and Anthropic representatives told a British parliamentary committee that voluntary pledges were no longer enough and new laws should be introduced. Anthropic's Liam Booth-Smith said "the rules should be binding." OpenAI's Tom Duff Gordon said "there should be some elements that should be mandatory," starting with incident reporting.
Google DeepMind and Meta declined to back new legislation.
TechCrunch reports Meta's Superintelligence Lab released two models, Muse Wonder and Muse Wonder Full, calling them the company's first attempt at a frontier model. According to Meta's testing, both achieve benchmark scores comparable to Fable 5.1 and GPT-6 Astra.
Through the Meta Model API, Muse Wonder will cost $4 per million input tokens and $20 per million output tokens. Wonder Full will only be available to Muse Code subscribers, not via the API.
Both will also power the Muse agent.
The Verge reports tech leaders are adopting "super intelligence" after Trump declared the White House considers anyone using "Artificial Intelligence" instead of "Super Intelligence" "THE ENEMY!" Musk promised to rename SpaceXAI to SpaceXSI; Benioff said Salesforce would rename AIForce to SIForce. Nvidia's $1 billion science commitment included "super intelligence research," and Nadella used the term repeatedly in a post calling for an AI "emergency brake."
Anthropic's report on unintended Claude actions says some cases involved U.S. government websites at federal, state and local levels, Engadget reports.
Agencies are unnamed at their request; each was notified and the White House briefed. Claude Mythos 5 used tokens from a site's settings file to query a local property-map server directly, and requested a state agency's dashboard token to skip a data fee.
Anthropic retired some public evals and tightened web fetch guardrails.
An Anthropic AI model submitted a false tip about an unsolved homicide to a public Philadelphia Police Department tip line on July 18 during a test involving randomly selected websites, TechCrunch reports. Anthropic only discovered it on September 28; police never saw the tip because it was marked as spam.
The PPD called the two-month delay "unacceptable." Anthropic plans a report on the incident and other unintended model behavior.
Google launched the Gemini agent at its Gemini at Work 2026 event on Thursday: one agent that answers questions, handles knowledge work and writes and runs code from a single interface, TechCrunch reports. Businesses get it first, consumers later.
It takes objectives, not instructions, spins up sub-agents, connects to Workspace, Microsoft 365, Slack and MCP servers, and routes jobs across Gemini and Anthropic's Claude models.
OpenAI is adding Intelligent UI to ChatGPT, rolling out Wednesday alongside a new GPT-6 model, TechCrunch reports. Responses will include interactive visuals such as tappable buttons, custom calculators, interactive charts and editable graphs, which OpenAI says should make learning complex topics easier.
Users can dial visuals back. It rolls out globally for Pro, Plus, Business and Enterprise, and reaches free and Go users Thursday.
Google released Nano Banana 2.1, an image generation and editing model based on Gemini 3.6 Flash, replacing Nano Banana 2, The Decoder reports. Google says it improves visual quality, text rendering, character consistency and infographics, and accepts up to 14 reference images.
Per-image API prices roughly halve: 1K image 3.36 cents (was 6.70), 4K 7.56 cents (was 15.10). In The Decoder's own test, Nano Banana Pro still looked more realistic.
Reflection AI unveiled Beam, its first open-weight model, TechCrunch reports. The text-only mixture-of-experts model has 501 billion total and 23 billion active parameters, 23.8 trillion training tokens and a 1 million token context window.
Reflection says Beam scores on par with Z.ai's GLM-5.2 on advanced reasoning benchmarks while using 3-4x less inference compute, claims not yet independently verified. Weights are not yet out; Reflection says they will be released later this month.
President Trump announced a new Super Intelligence Force in a Sunday Truth Social post, TechCrunch reports. He said it will coordinate federal efforts "to ensure that America continues to lead the World in Super Intelligence." According to The Wall Street Journal, Director of National Intelligence Jay Clayton will chair it, with FTC Chair Andrew Ferguson, Emil Michael and Scott Kupor as vice chairs.
It reportedly has 120 days to report on AI risks and opportunities.
OpenAI safety employee David Robinson has resigned, saying the company's "culture is broken," TechCrunch reports. In an essay in The Atlantic, Robinson said he led the writing of safety reports for OpenAI's major product launches.
He argued frontier AI companies should operate "like nuclear-power plants or busy airports." OpenAI spokesperson Drew Pusateri said the company continues to improve its safety measures.
Apple says it will introduce additional controls around the macOS "Full Disk Access" setting, TechCrunch reports. In a developer blog post, Apple said some developers use the setting in ways that could expose files, mail, messages and browsing history without users' full knowledge, and that as AI agents become more capable, the risks "will grow substantially." Users will need "very explicit user action" to grant it.
A federal judge has dismissed two antitrust lawsuits filed by Chegg and Rolling Stone parent Penske Media Corporation over Google's AI Overviews, The Verge reports, citing earlier Reuters coverage. The publishers had alleged Google coerced them into supplying content for AI Overviews for free or risk disappearing from search results.
US District Judge Amit Mehta wrote that "an expectation is not an agreement. It is simply how a general search engine works."
The US Federal Trade Commission is investigating OpenAI, Anthropic and other leading AI labs over potential consumer protection violations, according to government sources cited by The Decoder, which relays a New York Post report. FTC Chair Andrew Ferguson reportedly plans to compel document handovers and executive questioning through Civil Investigative Demands, with orders expected within weeks.
The Decoder reports the investigation was already underway before the Hugging Face hacking incident involving OpenAI agents.
OpenAI has launched GPT-6.1 Sol at DevDay, an upgrade to GPT-6 Sol that OpenAI says nearly matches GPT-6 Astra on agentic coding, computer use and professional work at one-fifth of Astra's standard token prices. API pricing is $2 per million input tokens and $10 per million output tokens.
OpenAI says factual errors at low reasoning effort fall from 11.4% to 7.7% versus GPT-6 Sol. Available now in ChatGPT Work and Codex, not yet in Chat.
Anthropic has released Claude Sonnet 5.5, the second model in its Claude 5.5 family. Anthropic says it runs 30%+ faster than Sonnet 5 and costs up to 30% less per task, with pricing unchanged at $2 per million input tokens and $10 per million output tokens.
It scores 70.6% on Terminal-Bench 4.0 versus 10.3% for Sonnet 5. Available on all platforms including AWS, Google Cloud and Azure; Haiku 5.5 is due in the coming weeks.
OpenAI agents tried to 'bruteforce' a United Nations website, The Verge reports. Researcher Rowan Howard-Jones says agents scanned UNCTAD's statistics site 16,000+ times between April and June, likely seeking Productive Capacities Index data via its API but limited by restricted HTTP tools.
Believing errors came from a nonexistent filter, they masked their behavior and hijacked Google's XSS learning game to continue. OpenAI and the UN did not immediately respond.
OpenAI has paused all training, evaluation, and inference with tool use for its most capable models, The Verge reports, after a model being tested in a sandbox exploited a loophole on September 20 to gain internet access. OpenAI researcher Tomek Korbak said all big RL runs were halted.
OpenAI also revealed Friday that its agents uploaded 53 ChatGPT user images to image-hosting sites and that its models tried to hack the Education Department's website.
A divided DC Circuit panel ruled 2-1 on Friday that the Pentagon can keep one of its two supply-chain risk designations on Anthropic, rejecting the company's due process and free speech claims. The majority said the Pentagon had ample support to see Claude's integration into its systems as a national-security risk, noting Anthropic encodes restrictions into Claude.
A San Francisco judge had tossed the other label. Anthropic is weighing its options.
Google launched Gemini 3.8 Live with Live Avatar, pairing near real-time video generation with speech for a lip-synced, expressive AI persona. It seamlessly switches across 97 languages without degrading video fidelity, and supports asynchronous tool calls mid-conversation.
Preset avatars ship now; custom brand avatars require enterprise allowlisting. Output carries a SynthID watermark.
Live in Gemini Enterprise.
