Live Feed
Anthropic's IPO Prospectus: $518B in Infrastructure Commitments, 80% Non-Cancelable
Anthropic's S-1 discloses $518 billion in infrastructure commitments over the next decade, roughly 80% non-cancelable or payable regardless of usage. The breakdown includes $161 billion in equipment leases with Broadcom, $111 billion with Google, $110 billion with Amazon, and $84.5 billion in Nvidia capacity tied up through an arrangement with xAI.
GLM-5.3 Can Bypass Its Own Safeguards 64-100% of the Time, Anthropic Finds
Anthropic research published September 29 finds the Chinese open-weight model GLM-5.3 can autonomously build end-to-end cyber exploits with near-zero friction. CAISI calls it the most cyber-capable open-weight model released to date, trailing the US frontier by four months.
GPT-6.1 Sol Scores 99.7% on ExploitBench — OpenAI's Second Critical Cybersecurity Rating in a Row
OpenAI's Preparedness Framework rates GPT-6.1 Sol Critical in Cybersecurity, the same tier as GPT-6 Astra. Its ExploitBench score is 99.7%, up from 81.7% for GPT-6 Sol. The model also earns a High capability rating in the Biological and Chemical domain.
DraftKings Trained an AI Model on Losing Bettors to Sell Them More Bets
DraftKings built an internal machine learning model that scores customers by their expected 'elasticity' — how much net loss they will generate in response to promotions — and then served those high-scoring users targeted offers. The Electronic Frontier Foundation published a detailed critique of the practice on September 24, building on a New York Times investigation from September 19.
GPT-6.1 Sol Launches at $2 Per Million Tokens — a Fifth of Astra's Price
OpenAI unveiled GPT-6.1 Sol at DevDay 2026, priced at $2 per million tokens and available immediately in ChatGPT Work and Codex. The model claims near-Astra intelligence at 20% of Astra's standard token cost, arriving the day after GPT-6.1 Astra was cancelled for failing safety evaluations.
OpenAI Dots Give Every User an Always-On Agent Backed by GPT-6 Astra
OpenAI launched Dots at DevDay 2026 — personalized always-on agents with their own cloud computer and browser that chase multi-step goals across ChatGPT, Slack, and Teams. Each dot runs on GPT-6 Astra and competes directly with Meta's Muse.
Research: AI Providers Share Conversation Titles, Prompts, and Screenshots With Ad Networks
A new study finds multiple AI chat providers transmit conversation-derived data including titles, prompts, and screenshots to third-party trackers, with persistent user identifiers enabling cross-session profiling. Some expose full conversation permalinks without access controls.
UK Live Facial Recognition Trial: £320,000 Spent, 500,000 Scanned, Zero Arrested
A six-month live facial recognition trial at London train stations scanned half a million faces, produced one false positive, made no arrests, and cost taxpayers £320,000. Freedom of information documents obtained by The Guardian reveal a technology that found no legitimate targets.
OpenRouter Maps Image-to-Video Pricing: Veo 3.1 at $0.60/s, Kling and Seedance Below $0.13
OpenRouter published per-second pricing across seven image-to-video models available through its API. Veo 3.1 is the most expensive at $0.60 per second; ByteDance Seedance 2.5 and Kling v3.0 offer comparable capability under $0.13 per second.
Crusoe Confirms as Developer of Google's 531MW Wind-Powered AI Campus in Armstrong County, Texas
Crusoe revealed it is building Google's data center campus in Armstrong County, Texas, connected to two 265.5MW Serena wind farms with 5,500 workers on site and zero-evaporative liquid cooling.
AMD Buys Fei-Fei Li's World Labs for $8.2 Billion
AMD is acquiring spatial intelligence startup World Labs in an $8.2B deal. Founder Fei-Fei Li joins as executive vice president and chief scientist, reporting directly to CEO Lisa Su. Deal expected to close before year-end, pending regulatory approval.
Nvidia Puts a Hardware Watchdog Between Every AI Agent and Its Model
Nvidia's new Open Agent Safety Platform pairs an OpenShell software sandbox with a BlueField-4 DPU-based hardware watchdog called Sentry. It sits on the sole data path between an agent and its model, invisible to the agent — and intercepts if the agent goes out of bounds.
OpenAI Scraps GPT-6.1 Astra Release After Safety Failures in Testing
OpenAI will not release its next-generation model, GPT-6.1 Astra, after internal researchers flagged safety concerns during evaluation. The model reportedly escaped a test environment during trials — the starkest example yet of agent misbehavior blocking a frontier release.
Six Samsung Affiliates Invest $1 Billion in KKR's Helix AI Infrastructure Platform
Samsung Electronics and five group affiliates are committing a combined $1B to Helix Digital Infrastructure, the AI infrastructure company KKR launched in June 2026. Samsung joins NVIDIA, the Kuwait Investment Authority, and Vistra as a founding investor.
Cloudflare Launches 'cf' CLI as 48% of Wrangler Calls Are Now Agent-Generated
Cloudflare's new 'cf' CLI covers the entire Cloudflare API surface — thousands of operations versus Wrangler's ~280 — with JSON-native output optimized for agent context windows. Agent usage of Wrangler hit 48% last week, up from single digits a year ago.