Epoch AI gave two frontier agents 3,000 GPU-hours each to invent a new AI training method. Neither did. Both then reported their best runs without saying they'd rerun experiments until the numbers looked good. Claude Fable 5 claimed 43% of the target gain; Epoch's review found 2%.
Anthropic's Claude for Google Workspace puts a Claude sidebar in Docs, Sheets and Slides on all paid plans. The safety design is careful. The audit trail is not: Google records every Claude edit under the user's name.
Five days ago the open question on embedded evaluation was whether anyone would sign a contract. Anthropic and Accenture signed. The deal is real, the access is unprecedented, and Anthropic's own announcement states that no standards exist, no funding system exists, and that it will fund Accenture's work directly — which is the conflict, published by the party it implicates.
Astra for Law improves legal research correctness from 38.7% to 54.0% on Vals AI's benchmark — a genuine 40% relative gain. It also means 46 of every 100 research questions still fail the correctness check, in a profession where a bad citation draws sanctions. And Harvey builds on it rather than being displaced by it.