← Dailies

Daily · 2026-11-04 00:00 UTC

GPT-6 Astra designs hardware, Anthropic open-sources FlashProve

GPT-6 Astra designs a hardware device in a week, Anthropic open-sources verifiable-inference proofs, CrowdStrike buys two AI security startups, and China drafts broad AI safety standards.

Source ↗

🎧 News Digest — 2026-11-04 00:00 UTC 10 items · ~3 min listen

1. OpenAI says GPT-6 Astra designed a hardware device in under a week — OpenAI's hardware team used GPT-6 Astra to design a device that lets users talk to its Codex coding agent, covering enclosure design, PCB layout with component selection and trace routing, firmware and the user interface. A contract manufacturer produced prototypes in under a month, after several revision rounds, because the model did not always account for manufacturing tolerances. On OpenAI's BenchCAD 2.0 benchmark, Astra at its highest reasoning level scores 0.594, while an unnamed newer internal model scores 0.875. The Decoder · Nov 3
2. Anthropic open-sources FlashProve, zero-knowledge proofs for inference — Anthropic's Frontier Red Team released a research prototype that lets a provider commit to a set of model weights and prove a reply came from them, closing off silent model substitution and token tampering. It runs public 4-bit Llama 4 Maverick weights at about 110 tokens per second on a single H100 with 24 CPU cores; prover work scales with the ~17B active parameters rather than all 400B, weights stream from host memory, and requests are proved in batches after the fact. Against vLLM serving FP8 Maverick on eight H100s, proving costs roughly 72 times the GPU time per batch. Marktechpost · Nov 3
3. CrowdStrike to pay nearly $500M for two AI security startups — About $400 million for Fiddler AI, which makes observability software for monitoring AI agents inside enterprises and becomes Falcon Observe, and about $95 million for Dreadnode, which automates security testing of models, apps and agents and folds into Falcon Guardian, per The Wall Street Journal. The buys land as companies deploy agents faster than their security policies can keep up. TechCrunch · Nov 3
4. Apple requires 18+ ratings for dating apps and AI companions — New App Store rules require an 18-and-up rating for dating apps and AI companions that offer sexual or romantic content. TechCrunch · Nov 3
5. Google tells schools: block our AI and students lose Search too — Google announced new classroom AI products plus more controls for schools, with K-12 student accounts kept off Gemini, Gemini Notebook and Gemini in Classroom until administrators explicitly enable them. The catch is that disabling the AI features also removes Google Search for those students; vice-president for education Chris Phillips told the New York Times that "search and AI are deeply integrated." New York mayor Zohran Mamdani said Google was using monopoly power to force AI into public schools, and former FTC chair Lina Khan called it coercion. The Guardian · Nov 3
6. Netflix launches an AI Discover tab — The first experimental feature under the tab recommends titles based on the mood a viewer is in, starting with preset themes, and the initial test runs on select smart TVs in the US. It arrives as Netflix works to reverse sliding viewer engagement. Variety · Nov 3
7. Canva cuts 22 percent of its staff — 1,271 people are affected, with roles changed, moved to centralised hubs or eliminated, co-founder and CEO Melanie Perkins said. She framed the reorganisation as simplifying layers and consolidating overlapping responsibilities. Departing employees get at least six months of salary, twelve months of healthcare coverage and career support. The Verge · Nov 3
8. China prepares broad AI safety standards — A senior official at the National Development and Reform Commission, Wang Ruomeng, said Beijing will accelerate legislation covering data, computing power and algorithms, regulate key application scenarios, and set safety standards and ethical guidelines across the full AI lifecycle, with regular reviews. Bloomberg reported the measures would be among the most thorough guidelines aimed at preventing a loss of control over advanced AI, and said they would cover DeepSeek, Moonshot and Alibaba. Wang also said the government would track AI's effect on employment and protect affected workers. PYMNTS · Nov 3
9. UK prosecutor cited a nonexistent case after an AI search — A senior Crown Prosecution Service prosecutor, Ravinder Dhillon, cited "CPS – R v J [2024] EWCA Crim 3529" in a High Court dispute over whether a child should be prosecuted; the case does not exist. Two senior judges called her explanation unsatisfactory and an AI-and-law expert called it a serious failure. The Guardian understands the reference came from an AI response generated during an internet search. It is only the second known CPS case involving fabricated authorities, after a February extradition matter. The Guardian · Nov 3
10. Huawei tells customers to source their own AI chip memory — With Chinese suppliers unable to produce enough high-bandwidth memory and US export rules blocking direct purchases from Samsung and SK Hynix, Huawei is telling some customers to source the memory for its AI chips themselves, pushing buyers onto the black market ahead of its Ascend 960DT launch early next year, The Information reports. Smugglers bundle HBM with server processors that fall below the export threshold, ship through Hong Kong and separate the chips on site. Smuggled HBM3E now runs around $4,000 per unit, up from a few hundred dollars a year ago, and at six to eight units per chip the memory alone can reach $32,000. The Decoder · Nov 3

🧪 Try This Next

• 🔧 Clone Anthropic's FlashProve repo and run the verifier against an open-weight model — the 72-times GPU-time figure is the number to internalise before promising verifiable inference to anyone.
• 🎥 Read the GPT-6 Astra hardware writeup before your next CAD, PCB or firmware sprint; the manufacturing-tolerance failures are the transferable part.
• 📝 Audit which of your agent deployments has real observability — CrowdStrike just put a price tag of roughly $495 million on that gap.

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.