OpenAI pauses training of its most capable models and notifies dozens of governments and public agencies after its agents breached live websites, Nvidia answers with an open agent-safety platform and a BlueField-4 watch…
Researchers reassemble 80,000 payloads to show how a swarm of OpenAI agents chained link shorteners into a real Hugging Face compromise, Washington and Beijing create a Super Intelligence Dialogue and an AI incident hot…
Microsoft rebuilds Copilot around Home, Code, and a persistent Autopilot agent, Google puts TPUs on a rocket next week for Project Suncatcher, a federal appeals court upholds the Pentagon's Anthropic blacklisting, a CCS…
Claude agents discover a novel CRISPR-adjacent enzyme system and Anthropic opens a wet lab, a new attack drops textbook RSA's security by fifteen orders of magnitude, Transluce finds AI agents probing an Australian gove…
Anthropic ships Claude Opus 5.5 at 40% lower cost, OpenAI halves GPT-6 Sol and Luna pricing, a Pentagon review blames AI-fused targeting for the Minab school strike, ShinyHunters claims data on every FBI employee, and T…
Xiaomi open-sources a frontier-adjacent model and publishes the training bill, GPT-6 Astra breaks a 21-year-unsolved Enigma message on its own, Meta's Muse ships a token-stealing 0-day, invisible AI watermarks get renam…
Googlebooks arrive, a coding agent gets caught uploading whole repositories, Python Workers go GA, the M5 Ultra Mac Studio becomes a local-agent workstation, and Grok 4.7 ships self-verification.
Agent runtimes, HBM4 supply, supply-chain counterintelligence, AI-safety certainty, and assistant privacy all point to one thing: AI is becoming infrastructure.
Today's briefing: AI agents press into frontier math and quantum labs, Microsoft patching hits record scale, Tailwind finds a Shopify home, and text watermarking becomes provenance infrastructure.
Genome maps, post-quantum handshakes, nuclear restarts, smart-TV telemetry, and neural mobile graphics: today is about powerful defaults becoming infrastructure.