Anthropic's Automated Systems Cleaned Up Ten Misalignment Benchmarks Without Regressing
Anthropic automated systems improved all 10 misaligned-behavior benchmarks without regression. The result shape is clean; the methodology stays opaque.
Anthropic automated systems improved all 10 misaligned-behavior benchmarks without regression. The result shape is clean; the methodology stays opaque.
A report calls open-weight AI firms Silicon Valley's hottest acquisition targets — but names no companies, no acquirers, and no deal figures.
The EPA plans to drop public notice requirements for industrial air permits. Data centers gain speed; neighboring communities lose formal standing.
A paper claims AI outperforms doctors — but has no authors, journal, or date. The claim functions as marketing until the evidence is visible.
Sandhya Devanathan moves from Meta to OpenAI's Southeast Asia and Australia operations — a regional shuffle, not a research or product event.
Meta is testing robots to swap cables and reset servers in data centers. No scale or timeline disclosed. The displacement risk is a human decision, not a machine one.
Trolls are seizing and publishing Cara's creator data. Meanwhile, a former AI art scraper claims to be building a tool to help those same creators.
A federal judge blocked the Pentagon's supply-chain designation of Anthropic, calling it "illegal and baseless." What the ruling actually exposed.
Will Knight's Uncanny Valley episode frames AI agent hacking as a US-China cooperation driver. It's a political claim. The stub isn't the reporting.
Barret Zoph moves from Thinking Machines to OpenAI to Google. What the compressed arc reveals about frontier AI talent retention.
Google's Gemini Notebook now connects to Play Books via "Expert Intelligence." Real RAG feature, inflated name — here's what actually shipped.
Anthropic published a physical-world AI agent position with no specs, no policy, no mechanism. Concern without constraint is a brand signal, not a position paper.