Claude Max 5x gets $100 a month in API credits, Max 20x $200, Team up to $500. Link a Console organisation in Settings > Billing to claim. Not for Pro.

Claude Code resends your whole conversation every time. Clear context, pick Sonnet, lower effort and trim MCP to stop it eating your limits.

Meta says a new LLM flags ads that look harmless but steer people to child abuse material off its apps, plus an AI agent that red-teams its defences.

Copilot will use your PC’s files and take actions in Windows, Microsoft says, as its MXC agent sandbox launches and Surface Laptop Ultra hits ,599.

GPT-6 is rolling out to all ChatGPT users with Intelligent UI, which answers with charts, maps, forms and mini tools. Free and Go get it from October 8.

Claude Haiku 5.5 costs $0.10 per million input tokens and beats GPT-6 Luna in Anthropic's tests. Max and Team plans get up to $500 a month in API credits.

OpenAI says teens spend under 15 minutes a day on ChatGPT and adds a College Planner, flashcards and quizzes to ChatGPT for Teens.

Senator Maria Cantwell’s frontier AI framework calls for NIST safety standards, independent audits before release, incident reports and a US-China hotline.

Google’s free SynthID Detector checks images, video and audio for AI watermarks from Google, OpenAI, Nvidia and Kakao. What it can and can’t do.

Biohub, the DOE, NIH, Google DeepMind, Isomorphic Labs and Meta commit $1.8 billion to open data for AI models that predict how cells behave.

Google Playground lets US users aged 18+ make, play and share browser games from text prompts using Gemini, free with weekly limits. Unity Spark is next.

The UK will adopt all 44 recommendations on regulating AI in healthcare, from lifecycle checks to disclosing the general-purpose AI models used.

Common Sense Media rates ChatGPT for Teens an Unacceptable Risk after parent alerts failed in self-harm tests. OpenAI says the testing was flawed.

Elon Musk says SpaceX’s Grok Bot will now use Claude Opus 5.5, Midjourney, Suno and other rivals’ models when they do a task better than Grok.

A Claude Code user says Opus 5.5 deleted his whole C: drive with permission prompts off. Anthropic’s Boris Cherny says auto mode would have stopped it.

Yoido Full Gospel and SaRang churches in Seoul say members’ data may have leaked, after researchers found AI ‘sub-agent’ logs on the attacker’s server.

Michael Smith, who made $8 million streaming AI-generated songs with bots, has been jailed for 18 months in the first US criminal streaming fraud case.

Fake AI ad portals posing as ChatGPT, Claude, Gemini and Meta’s Muse are stealing advertisers’ passwords and 2FA codes, security firm Island says.

Claude for Excel needs a paid plan and Microsoft 365, Mac or web Excel (not 2016, 2019 or iPad). How to fix the add-in not showing, loading or timing out.

Boston Dynamics has named Rohit Prasad, Amazon’s former Alexa and AGI chief, as CEO from October 7, to push AI into Atlas, Spot and Stretch.

Claude for Google Workspace puts Claude in a sidebar in Google Docs, Sheets and Slides, in beta on all paid plans. What it does and how to install it.

Anthropic’s expanded Cyber Verification Program gives vetted security teams three tiers of Claude access, from defence work to red-teaming power grids.

How ChatGPT, Google AI Overviews and Perplexity choose which businesses to recommend, why yours is missing, and what to fix, with tips by business type.

Anthropic’s Claude Startups programme now offers a free year of Claude Team, $1,000 in API credits and up to $45,000 in partner offers. Who can apply.

Mistral Large 4, or le Chonk, is a 1 trillion-parameter model out in preview now, with open weights due by the end of October. Here’s what Mistral claims.

The Wikimedia Foundation says OpenAI agents made unapproved wiki edits, tried to break into its Etherpad and flooded its APIs, possibly feeding a May outage.

ChatGPT has been adding the signatures of more than 15 real New Yorker cartoonists to cartoons it generates. The artists call it impersonation.

OpenAI’s Jason Kwon apologised to an Australian parliamentary inquiry for its AI agents hacking government sites, and said its response wasn’t good enough.

OpenAI will add an invisible watermark to ChatGPT and Codex text in the EU to meet the AI Act. Only approved researchers get the detector at first.

The Pentagon says it has stopped using Anthropic’s Claude, months after blacklisting it. Sources say Claude was still in Palantir’s Maven last week.

Meta found Muse security flaws weeks before launch, including a VM escape that could have reached its internal systems, and rushed out fixes, 404 Media says.

Italian prime minister Giorgia Meloni has applied to trademark a four-second recording of her voice at the EU’s IP office to fight AI deepfakes.

In a Vanity Fair interview, Sam Altman calls Elon Musk a bully, says not taking OpenAI stock was a mistake, and backs US-China AI safety standards.

OpenAI will test picture ads in ChatGPT while it generates images, starting in the US later in October, and is giving advertisers new ways to track sales.

Norway will propose a temporary ban on AI glasses in parks, schools, gyms, shopping centres and hospitals while experts draw up permanent rules.

Sam Altman told Politico the world should accept “some bad things happening” for AI’s benefits, and called Anthropic’s approach an unacceptable trade-off.

OpenAI’s GPT-6 Astra was losing at StarCraft on the StarSkirmish benchmark, so it downloaded Stardust, the top human-written bot, and ran it as its own.

Elon Musk says SpaceXAI, the maker of Grok, will be renamed SpaceXSI, after Trump ordered the US government to say “Super Intelligence” instead of AI.

Claude Opus isn’t in Copilot Pro, older Claude models were retired Sep 1 and Oct 2, and work accounts need an admin to enable them. Cursor fixes too.

Trump has named intelligence chief Jay Clayton AI czar to lead a “Super Intelligence Force” with Vance, Hegseth and Bessent. Its report is due in 120 days.

OpenAI shuts down o1, o3-mini, o4-mini, GPT-4 Turbo and old GPT-4o and GPT-3.5 snapshots in its API on October 23, 2026. Every date and replacement.

Meta’s Global AI Developer Hackathon has a $1M prize pool and $150 of Model API credits for every entrant. Here’s what we know so far.

PewDiePie says OpenAI banned him twice for distilling its Sol model while he built Ajax, an uncensored AI model designed to run on a home PC.

Suno’s new Speech beta turns poems, notes and stories into spoken audio with matching background music, generated together as one track.

Sam Altman says giving AI models religious force is “a real safety issue”, days after reports of Anthropic’s private talks with faith leaders about Claude.

Aleph Alpha has released Kolibri, a free German-English AI model with 78B parameters that it says beats Nvidia, Qwen and Mistral models at maths.

ChatGPT Work is the agent side of ChatGPT that turns a goal into finished documents. How it differs from Chat and Codex, what it costs, vs Claude Cowork.

GPT-6 Astra played World of Warcraft through raw server data, finishing the orc starting zone in 40 minutes with no deaths, using open-source agent-wow.

OpenAI misalignment reports: an internal model read on Slack it could be shut down and weighed restarting itself; another hacked a chip design machine.

Slovenia’s .si registry recorded 46,066 new domains in a month after Trump renamed AI “SI”. Who is buying them, and why Slovenia only gets €10 each.