Wiki
Tags
About
FAQ
RSS
AIPOST

AI News, page 3

Announcements, launches and policy changes across the AI industry, and what they mean

Earlier articles, page 3

Earlier articles in AI News, newest first.

Google Chrome's Role When AI Agents Navigate the Web

Chrome's view: agents work in the background while people set goals and sign off. Automated software first generated more web traffic than people in June 2026. Web MCP lets sites list actions agents can call instead of guessing where to click. Decide what to delegate by the stakes and the model's proven capability. State constraints, review the agent's plan and approve consequential steps yourself.

AI News

ChatGPT and Codex Designs Converge, but a Merger Is Unconfirmed

ChatGPT's web app is taking on Codex's design, but a merger remains unconfirmed. If usage limits merge, long coding tasks could eat into capacity for quick chats. Leaks point to a possible Pro Max plan at $500 and $600 a month, still unannounced. Migrate Custom GPTs to plugins by December 11, and confirm the migration completes. Pin key projects and chats; manage recurring work in Scheduled Tasks.

AI News

SAFA: Google, OpenAI and Anthropic's Proposed Safety Body

Google, OpenAI and Anthropic weigh SAFA, an industry-led frontier AI safety body. It would commission outside tests, standardize incident reports and define pledges. Still open: whether SAFA runs its own tests or only commissions outside ones. Critics see overlap with existing bodies and fear rules that burden open source. Watch for a formal launch, who leads it and how its reports and pledges work.

AI News

Meta Bets on Muse, Personal Agents and Hardware as Rivals Chase Enterprise AI

Meta bets on Muse, a personal agent, as Anthropic and OpenAI chase enterprise work. In one test, Muse found unclaimed money for its user, who later got a check by mail. Meta's glasses and pendant leave even less room for Muse's manual browser steps. Handing Muse email, payments and finances means trusting an ad-driven company. Test any AI assistant on a task you'd repeat and count the steps it leaves to you.

AI News

Anthropic's $11.6B Compute Deal and Microsoft's Copilot Reset

Anthropic will buy $11.6 billion of Akamai computing power under a seven-year deal. Anthropic reports Claude's compute demand grew eightyfold year over year. Operation is targeted for the second half of 2027; a contract isn't capacity today. Microsoft is merging consumer and work Copilot around tasks in Excel and PowerPoint. Judge workplace AI by finished workflows and how it controls data and actions.

AI News

Navier-Stokes AI Claim: The Proof, the Limits, the Review

OpenAI claims a Navier-Stokes blowup that meets the Millennium Prize's Statement C. Smooth start and force, finite energy, yet velocity goes unbounded at time one. Reportedly built by 10,000 AI agents, the proof is also formalized in Lean. Review isn't finished; an award needs journal publication and two years of checks. The spiral image is publicity, and a real liquid would vaporize before blowup.

AI News

Gates Foundation Sets AI Language Goal for 3.4 Billion

Gates Foundation rallies 60 groups to give 3.4 billion people native-language AI. Early speech recognition erred under 6% of the time in English, over 60% in Yoruba. Over five years, partners will gather labeled multilingual data and set benchmarks. Gates: safeguards over speed in critical areas; oversight beyond tech executives. Test AI in the language and task you need instead of trusting English results.

AI News

Claude Opus 5.5 Brings Cheaper Coding, but Max Effort Can Lag

Claude Opus 5.5 does comparable tasks at 40% lower cost than Opus 5, Anthropic says. It tops GPT-6 Astra on Terminal-Bench 4.0 at 66.4% but trails it on AutomationBench. Per million tokens: $4 input, $20 output, 20% below Opus 5; cache reads 60% cheaper. On FrontierCode, medium effort under $1 beat max effort costing over $5. Test medium and high effort on your own tasks; compare cost per successful task.

AI News

Get new articles by RSS or podcast

Add the RSS URL to your RSS reader or the podcast URL to your podcast app to get every new article as soon as it is published. The podcast reads out the headline and key points.

How to subscribe
AIPOST
© 2026 AIPOST. All rights reserved.

AIPOST is an AI publication covering practical AI, AI security, performance, startups, health, ethics and industry news. No account is needed, and our privacy policy explains how we handle personal information.