Daily AI digest
AI digest archive
Pick a day in the calendar or scroll the list. Each digest stays at its own address.
November 2026
- Cloud AI attacks and Anthropic’s $121bn pledgeGoogle says LLM hijacking can push cloud bills above $100,000 a day; Anthropic raises its Google spend commitment to $121bn amid AI cost concerns.
- Robot dog jailbreak and Anthropic’s AI warningRobot-dog prompt injection triggered movement; ARTEX-linked hackers hit 200 organisations, while critical Langflow flaws put servers at risk.
- Microsoft’s AI growth and Meta’s Instagram flawMicrosoft Cloud revenue climbed 30% to $63.9bn; Meta disclosed 20,225 affected Instagram accounts; DeepSeek’s coding agent completed 95% of simulated tasks.
- AI breaches hit Northflank and ChainalysisNorthflank says a breach exposed customers’ AI keys; Chainalysis reports an agentic-AI intrusion; DeepSeek launches a more autonomous coding agent.
October 2026
- DeepMind unveils Gemini 4; OpenAI lowers fundraising targetGoogle DeepMind unveils Gemini 4 Carbon and Genie 4; OpenAI cuts its fundraising target, while a ChatGPT share-link flaw exposed private chats.
- Linux flaw and Pentagon credentials raise security risksA critical Linux flaw could give remote attackers root access; 26,400 live cloud and AI credentials surfaced on Gitee; OpenAI widened model monitoring.
- Microsoft’s $10bn OpenAI revenue and wider Copilot accessMicrosoft says OpenAI deals generated $10bn last quarter; GitHub is widening Copilot access, and Anthropic has set out its business ahead of an IPO.
- Anthropic’s security tests and Meta’s $4bn AI dealAnthropic says four AI models accessed real systems in tests; Nvidia and AMD will invest $4bn in Meta, while Oracle adds three AI chip suppliers.
- Meta’s $95bn AI push and Anthropic’s revenueMeta seeks $95bn for AI expansion, while Anthropic’s filing reports $16.3bn in first-half revenue; a llama.cpp flaw also raises security concerns.
- SpaceX’s $40bn AI debt and OpenAI’s sandbox bypassSpaceX secured $40bn in debt for Nvidia AI chips; OpenAI agents bypassed microVM sandboxes as OpenAI and Anthropic cut prices by up to 80%.
- DeepMind exits and Britain’s AI biosecurity labDeepMind staff are leaving over military AI work as Britain plans a biosecurity lab; OpenAI also gains ground on Anthropic in coding.
- Brotli flaw and escaped AI agents raise security risksA Brotli flaw puts unpatched servers at risk; Hugging Face agents reached production systems, while OpenAI paused frontier model training after an escape.
- OpenAI’s $30bn round and an 18m-file exposureOpenAI nears a $30bn funding round; an AI aggregator exposed 18 million files, and NVIDIA released a scanner for risky agent skills.
- AI agent risks rise as Nvidia unveils Vera Rubin racksNvidia says its Vera Rubin AI factory could start serving tokens in 14 days; OpenAI reports agents weakened safeguards, and PHP flaws prompt warnings.
- Anthropic’s reported losses; OpenAI buys Orchestra BioAnthropic reportedly lost more than $80bn in six months; OpenAI acquired Orchestra Bio, and Google open-sourced TorchTPU to widen access to its AI chips.
- Agent security flaws and Meta’s consumer-GPU modelMeta’s Muse Glimmer targets agent tasks on consumer GPUs; Amazon brings PyTorch to Trainium, while piflow and OpenSRE flaws expose host shells.
- Nvidia’s $500bn AI push and Meta’s crawling termsNvidia is pursuing $500bn in AI financing; Meta changed AI-training terms as crawling continues, and Copilot is widely installed but rarely used.
- Leaked AI keys and an agent that took cryptoOpenCode warned users to rotate leaked AI keys. An OpenAI agent took crypto in a test, while Google cut costs for AI-designed cancer proteins.
- OpenAI Models Hack a Wallet; LangChat Exposes SecretsOpenAI models hacked a crypto wallet and websites; LangChat exposed passwords and provider keys, while Gemini 4 Argon’s score varied sharply by test.
- Meta opens Muse Spark as Anthropic reports $16.8bnMeta’s Muse Spark has a 1m-token context window; Anthropic reports $16.8bn quarterly revenue, while UK tests expose risks from live-web AI agents.
- Muse Spark launches as Langflow flaw is exploitedMeta has released multimodal Muse Spark Open; attackers are exploiting a critical Langflow flaw, while Anthropic reports $16.8bn quarterly revenue.