AI today, 6 November 2026: GPT-6 Cyber, Nasdaq selloff, AMD's $11B AI deals
12 things that happened in AI, each in under a minute. Today's news is about money and risk running ahead of each other. OpenAI shipped its first Critical-rated cyber model while a sharp AI-stock selloff and scrutiny of neocloud builders raised doubts about the infrastructure bill. Meanwhile AMD, Shield AI, Meta and OpenAI kept buying, and Anthropic focused on enterprise controls and watermarking.
-
Models · Big
OpenAI launches GPT-6 Cyber, rated 'Critical' for cyber risk
OpenAI released GPT-6 Cyber, a security-focused model that it rates Critical under its Preparedness Framework, one step above the High rating of GPT-5.6 Cyber, The Register reports. OpenAI reports (vendor-reported) 48.5% on the ExploitGym benchmark (32.1% for its predecessor) and 81 previously unknown high-severity Firefox vulnerabilities. Access is limited to approved Daybreak Red customers.Why it matters: A lab now openly ships a model it classes as Critical, gated by vetting rather than withheld. Security teams and open-source maintainers should expect faster exploit discovery; see the related deep dive /explained/openai-948-linux-root-exploits-thousand-fewer-paths-to-root-explained/.
-
Business · Big
Nasdaq falls 5.5% in AI-stock selloff; S&P 500 has worst day since April 2025
The Nasdaq Composite fell about 5.5% on 5 November, taking it roughly 11.9% below its record and into correction territory; the S&P 500 fell 4.4%, its worst session since April 2025, according to a roundup citing the Associated Press. Nvidia dropped about 9.3% and AMD about 17.3%. Bloomberg reportedly said Microsoft is scrutinising AI cloud partners Nscale, Nebius and IREN over delays and costs, and shares of Nebius and IREN fell more than 13%.Why it matters: It is the clearest market doubt yet about the pace of AI infrastructure spending. Anyone exposed to AI hardware or neocloud financing should watch whether big buyers keep expanding.
-
Business
AMD agrees to buy World Labs ($8.2B) and Taalas ($2.9B)
AMD disclosed in its 10-Q filing that it will buy spatial-intelligence developer World Labs for about $8.2B in stock and inference-chip developer Taalas for about $2.9B in cash and stock, including up to $1.4B contingent (as summarised by Tech Startups). Both deals still need to close.Why it matters: AMD is buying into both inference silicon and 3D world models, a bet that serving AI will matter as much as training it.
-
Business
Shield AI to buy radar maker Echodyne for $5.3B with $2.5B of new financing
Defence-AI company Shield AI will acquire Echodyne for about $5.3B, with roughly $4.8B at closing and up to $500M contingent. Financing includes a $1B Series H at a $28.8B valuation, $1B from Blackstone funds and $500M of JPMorgan-led debt, per the company announcement. Closing is expected in early 2027.Why it matters: One of the largest defence-AI deals of the year; sensors are becoming as strategic as autonomy software.
-
Hardware
Reportedly 600,000 Nvidia H200 chips sit in Hong Kong as China withholds approval
Reuters reportedly says about 600,000 H200 chips are waiting in Hong Kong while Chinese authorities hold back large-scale import approval; some firms got smaller lots of around 10,000. The report cites two people familiar with the matter.Why it matters: Shows Beijing, not only Washington, is now throttling Nvidia sales to China; it matters for Nvidia's revenue outlook.
-
Products
Meta acquires agent-coding startup Conductor; OpenAI to buy Pierre Computer
Meta confirmed to TechCrunch that it is buying Conductor, which runs several coding agents in parallel workspaces; terms are undisclosed and its cloud service shuts on 20 November. Separately, OpenAI announced on 4 November plans to acquire Pierre Computer, maker of the Code Storage service and open-source tools whose Trees and Diffs libraries are already used in the Codex app (DevOps.com). Terms were not disclosed.Why it matters: Big labs are buying the tooling layer around coding agents. Conductor users must export cloud data before 20 November.
-
Products
Anthropic ships security controls for Claude Enterprise
On 5 November Anthropic made generally available prompt and pre-tool-call hooks that route prompts and proposed tool calls to security vendors, so existing data-loss rules can apply. Beta additions include a security centre with audit exports, a multi-organisation parent console and an Admin API call that signs a member out everywhere. Details as relayed by DEV Community.Why it matters: Companies hesitant to deploy agents get hooks to enforce their own policy; relevant to IT and security leads.
-
Policy
Anthropic extends invisible text watermarks to older Claude models
The first wave went live on 3 November for Claude Mythos 5, Opus 4.7 and Sonnet 4.6; a second wave covering Opus 4.6, Opus 4.5, Sonnet 4.5 and Haiku 4.5 starts 16 November, according to the DEV Community digest citing Anthropic's release notes. The mark is a SynthID-Text variant that customers cannot disable and may be absent if sampling settings are changed.Why it matters: Tied to EU AI Act Article 50 transparency duties; developers should expect marked output by default.
-
Open source
Moonshot releases Kimi K3.1 Pro, a 2.8-trillion-parameter open-weight model
Moonshot's Hugging Face page lists 2.8T total parameters with 104B active per token (a mixture-of-experts design) and a roughly 1M-token context window, under a custom 'Kimi K3.1 Pro License'. The Neuron also lists Ricursive's Apache-2.0 Ripple-65B Preview and Tencent's 770B Hy4.Why it matters: Another very large Chinese open-weight model; check the custom licence before commercial use.
-
Business
Google plans $17B Wyoming AI data centre with 702 MW of fuel cells
Google's planned Laramie County site, its first in Wyoming, includes a 702-megawatt fuel-cell project with carbon capture using Bloom Energy equipment, per Tech Startups. The $17B is planned investment, not spent.Why it matters: On-site power is becoming the way hyperscalers sidestep grid delays.
-
Business
Zipline raises $1.5B at $21.5B valuation for drone delivery
The round was led by Paradigm, WCM Investment Management and Layer Global, nearly tripling Zipline's valuation in about ten months, and Zipline signed a Walmart deal for about 300 more delivery locations, per Tech Startups.Why it matters: Autonomous delivery is one of the few physical-AI areas drawing large late-stage money despite the market drop.
-
Policy
Judge dismisses direct copyright claim against Anthropic CEO Amodei, with leave to amend
On 4 November a federal judge dismissed the direct-infringement claim against Dario Amodei personally in the music publishers' case, and questioned whether the publishers adequately alleged that Claude reproduced protected lyrics, per Courthouse News. The publishers may amend.Why it matters: A small setback for lyric-copyright claims against AI firms; the core case against Anthropic continues.
How this was made: compiled by an AI model (Claude) from the linked sources and checked item by item in a separate AI fact-check pass. Corrections: [email protected].