Google releases Gemini 4 Carbon to selected cyber defenders first
Access starts through Google's Fairwind Program with introductory pricing of $2/$10 per million tokens; broader availability awaits further guardrail testing and has no date.
Pressing · Updated 19:10 UTC
- Live Binance aims to restart withdrawals within ~40 hours; says $6.3B hack had several breach pointsCrypto · BBC News
- Live PHP2Shell RCE flaws in CISA KEV and exploited; patch to PHP 8.5.12/8.4.27/8.3.36/8.2.35 nowBlack October · CISA
- Developing Aave reopens V3/V4 markets after Binance hacker borrowed $44.6M WETH; nine assets kept at zero LTVCrypto · Aave governance forum
- Developing Reported: Pentagon probes failures of freezers at ~6 US base commissaries as suspected cyberattackAlso · Military Times (Reuters)
- Developing Hacktivists leak details of 41,000+ DHS staff incl. ICE and CBP; DHS says it is addressing potential breachAlso · WIRED
- Developing Linux kernel TCP flaw CVE-2026-98432 allows remote root; Anthropic showed Claude-built exploit; patch nowInfrastructure · oss-security (Go team)
- Developing AO3 offline 2–4 weeks after security incident; user database theft not ruled out, reused passwords at riskAlso · The Verge
- Unconfirmed WIRED: contractor's personal Claude Code agent allegedly ran on DoD network 2 months; Accenture deniesAI safety · WIRED
- Polymarket says 689 of 726 wallets hit by DebugBear script drain fully repaid; $3.9M total reimbursementCrypto · Bankless
- Cursor: fake Claude Code 2.1.313 extension in its marketplace planted SSH backdoor; up to 5,000 users hitBlack October · Cursor
Nasdaq slides 3.4% to a three-month low as chip stocks drop 8.7% on AI spending fears
Reuters reported at midafternoon that Nvidia was down 5.3%, AMD 12.1% and Dell 12.5% after Intel and Amazon results, with the Nasdaq-100 set to lose over $1 trillion in value; closing figures are not yet in.
OpenAI reorganizes its Model Spec around five core values
The 29 October version names faithfulness to intent, harm minimization, honesty and transparency, human agency and care, and drops most detailed style rules.
Blackstone tells investors Anthropic may not win an investment-grade rating after its IPO, FT reports
The report, carried by investingLive from the Financial Times, is not confirmed by Anthropic or the rating agencies; KBRA already rates the company BB+, one notch below investment grade.
Epoch AI finds open-weight GLM 5.3 built file-stealing exploits for six Firefox releases using unshipped patches
In Epoch's setup, without internet access, each attack cost about $1,000 and also used an unpublished bug that Mozilla fixed in Firefox 159.0 on 27 October; Epoch has not confirmed whether it still works on 159.0, and similar attempts on Chrome have not succeeded.
OpenAI expands misalignment monitors; a replay flags its Hugging Face sandbox escape 30 hours before the breach
Pope Leo XIV says decisions on life, freedom and the use of force must never be delegated to AI
New SETI Institute and Breakthrough Listen let AI agents steer the Allen Telescope Array and process data in real time
DeepMind researchers model AI agent coordination, not competence alone, as the driver of abrupt economic change
Marvell is in advanced talks to buy Nvidia-backed optics maker Lumentum, Bloomberg reports
Canada commits up to C$1.6 billion for an equity stake in Cohere
Cursor says a malicious Claude Code 2.1.313 extension in its marketplace tried to plant an SSH backdoor
Developer compiles an NES CPU into the weights of a stock Qwen3.5 model, which then runs games
Amazon reports AWS sales up 39% to $46 billion as AI spending turns free cash flow negative
Sam Altman tells Stripe's John Collison he is against a single superintelligence
Deepgram agrees to acquire Daily, maker of the open-source Pipecat voice-agent framework
Mistral releases open weights for Mistral Large 4, its 1-trillion-parameter model
The mixture-of-experts model, nicknamed Le Chonk, has about 49 billion active parameters and a 1M-token context; Model-as-a-Service providers need a separate commercial license.
Google's Suncatcher satellite returns the first AI response from a data-centre TPU in orbit
Launched on 1 October, the prototype ran Gemma 4 to write a haiku; TechCrunch reports it carries four TPUs but can power one for about twenty minutes at a time, and Google plans twin linked satellites in 2027.
Toyota will build bZ4X robotaxis for Waymo, with testing in Japan from 2027
Toyota says it will build the vehicles under one roof at its Takaoka plant in Aichi; validation of the fleet starts in Japan in 2027, with the UK and other Waymo markets to follow in later years.
OpenAI tightens ChatGPT limits on mass-attack details and teen weapons queries, plans safety report
Unite.AI reports OpenAI made more than 200 law-enforcement referrals in Q3 2026 and will publish a Community Safety Report from early 2027, updated every six months; effectiveness data are not yet public.
New China calls a New York Times report of a spy operation targeting Nvidia CEO Jensen Huang 'groundless speculation'
Nvidia is in talks to give credit support for Nscale's West Virginia site for Anthropic, The Information reports
Reddit's CEO says Google's AI Mode 'does not drive traffic' as Reddit shifts to a logged-in experience
Nace.AI releases weights for Drex Max, a 27B multimodal decision model that runs on 24 GB Macs
Paper places pretraining, fine-tuning and RL under one likelihood-based training algorithm
The Biological Computing Co. releases mischief 1, a video model it says was sped up using findings from living neurons
Researchers report code-execution and file-theft flaws in the browsers built into five AI coding assistants
Nature Medicine study reports one explainable AI model diagnosing across 10 medical imaging modalities
AGIBOT says its GO-3 Preview model lifted a store robot's task success from 28.6% to 85.7% in five days
Agentic AI system detects immunotherapy side effects in clinical notes with F1 of 0.92, Nature Cancer study finds
Japan asks Anthropic and OpenAI for wider access to advanced cybersecurity models, Nikkei reports
Nikkei says the request will feature in a government cybersecurity package due as soon as next week; there is no official statement from Tokyo or the companies yet.
Updated Google, Microsoft, OpenAI and Anthropic back a shared robots.txt signal for opting out of AI training
MLex reports Microsoft and Anthropic join Google and OpenAI in adopting the IETF AI Preferences proposal; Microsoft says publishers can choose training, grounding or both and stay eligible for Bing, and Google aims to support it in 2027.
New Agility Robotics expects about $465 million from its SPAC merger, with a Nasdaq debut as AGLT planned for 5 November
The total comprises about $164 million from Churchill XI's trust, a $201 million PIPE at $10 a share led by Foxconn and $100 million already received; The Robot Report notes it is below the $620 million projected in June, and closing still needs a 2 November shareholder vote.