Live Feed
vix Scaffold Puts Claude Opus 4.7 at 90.2% on Terminal-Bench 2.0 — 5.5 Points Clear of GPT-5.5
A third-party scaffold named vix running on Claude Opus 4.7 claimed the Terminal-Bench 2.0 top spot today at 90.2%, the first model to break 90% on the benchmark and a 5.5-point gap over the nearest GPT-5.5 entry. OpenAI's Codex CLI, which held the lead at 82.0% last month, now sits at rank 6.
Anthropic's 2028 Paper: Lock Down Compute Now or Risk Losing the AI Lead to China
Anthropic published a paper arguing the US and allies can secure a 12-to-24-month frontier AI lead by 2028 — but only if China's access to advanced chips and model outputs is closed before the window narrows. Huawei is currently producing 4% of NVIDIA's aggregate compute. Distillation, the paper argues, is systematic industrial espionage.
Microsoft Cancels Thousands of Claude Code Licenses by June 30 — Copilot CLI Takes Over the Agentic Stack
Microsoft's Experiences and Devices group — responsible for Windows, Microsoft 365, and Teams — has been ordered off Claude Code by June 30. EVP Rajesh Jha cited control over Microsoft's own repos and security requirements. Developer preference still runs toward Anthropic.
Mythos Chained 2 Unknown macOS Kernel Bugs Into a Working Exploit in 5 Days — Apple Is Investigating
Security researchers at Palo Alto firm Calif used early Mythos access in April to discover two unknown macOS kernel vulnerabilities and build a privilege escalation exploit that bypasses Apple's memory integrity protections. Apple has confirmed it is investigating.
146,932 Hallucinated Citations Entered the Scientific Record in 2025. Now Venues Are Fighting Back.
A new paper auditing 111 million references across arXiv, bioRxiv, SSRN, and PubMed Central finds a sharp post-2024 rise in AI-fabricated citations with no sign of plateauing. ICLR 2026 desk-rejected 647 submissions. ICML and ACM CCS have announced similar policies. arXiv's CS chair is publicly affirming author liability.
Ontario Audit: 12 of 20 Government-Approved AI Medical Scribes Got the Drug Wrong
Ontario's Auditor General found that most AI scribes approved for use by the province's doctors hallucinated treatment plans, misidentified drugs, and missed mental health details in standardised tests. The province weighted 'accuracy of medical notes' at 4% of the procurement score. Domestic presence in Ontario was weighted at 30%.
OpenAI Brings Codex to Mobile: 4 Million Weekly Users Can Now Direct AI Coding Agents From Their Phones
OpenAI launched Codex in the ChatGPT mobile app on May 14, giving developers remote access to running agents, live diffs, approval queues, and terminal output from any device. A secure relay layer keeps the host machine unreachable from the public internet.
Six State AGs Ask SEC to Probe Sam Altman's Investments Before OpenAI's IPO
Republican attorneys general from six states and the House Oversight Committee have opened parallel investigations into whether Sam Altman's personal stakes in Helion, Stoke Space, and other startups constitute conflicts of interest with OpenAI's commercial relationships, putting direct pressure on the company's IPO path.
Apple-OpenAI Partnership Fractures as OpenAI Explores Legal Action
Bloomberg reports the Apple-OpenAI relationship has deteriorated badly enough that OpenAI is now exploring legal options against Apple, as Gemini replaces ChatGPT as Siri's AI backbone and OpenAI builds competing consumer hardware with Jony Ive.
Anthropic Commits $200M With the Gates Foundation to Vaccines, Global Health, and K-12 Education
Anthropic and the Gates Foundation are deploying $200M over four years to accelerate vaccine discovery for polio and HPV, improve disease forecasting in low-income countries, and deliver AI tutoring to students in sub-Saharan Africa and India.
OpenRouter Ships Human-in-the-Loop Approval Gates and Session-ID Stickiness for Production Agents
OpenRouter's May 11 update adds two primitives production agent teams have been building in-house: HITL tool calls that pause an agent loop for human review, and session-id stickiness that routes multi-turn conversations to the same provider backend for better cache performance.
Apple Rebuilds Siri as an Always-On Agent for iOS 27, on a $1B Gemini Foundation
Apple is scrapping the legacy Siri architecture for iOS 27 and replacing it with a system-wide AI agent built on Google Gemini at $1B/year. A dedicated Siri app, Dynamic Island integration, and third-party agent extensions debut at WWDC 2026 on June 8.
Cerebras Prices IPO at $185, Raises $5.55B — Largest US Tech Debut Since Uber in 2019
AI chipmaker Cerebras priced its Nasdaq IPO at $185 per share on May 13, above the $150-160 marketed range, on 20x oversubscription. Fully diluted valuation: $56.4B. Ticker CBRS begins trading May 14.
Princeton Ends 133-Year Honor Code Precedent: AI Cheating Forces First Exam Proctoring Since 1893
Princeton faculty voted nearly unanimously on May 11 to require proctors for all in-person exams, ending a practice students won in 1893. Generative AI made cheating easier and harder to catch simultaneously. Effective July 1.
Anthropic Launches Claude for Small Business: 15 Agentic Workflows, 7 Connectors, No IT Required
Anthropic's fifth market-specific Claude bundle integrates with QuickBooks, PayPal, HubSpot, Canva, DocuSign, and Google and Microsoft Workspace in a single toggle install. Small businesses account for 44% of US GDP; most AI adoption had stopped at the chat window.