METR: Anthropic’s AI R&D is accelerating about 2×
Independent evaluators at METR say Anthropic is now improving model capabilities on well-specified tasks at roughly twice the rate seen between March 2024 and October 2025 — with high uncertainty (they would not be surprised by 1.3× or 3×). Benchmark scores via Anthropic’s AECI have grown at an estimated 1.9× the pre-October 2025 rate. A majority of that lift, METR judges, comes from extensive use of AI to do R&D rather than a jump in training-compute scaling. The catch for outsiders: Anthropic redacted most of the evidence METR found most informative, so the public must lean on METR’s judgment. No clear bottlenecks showed up for the next year, which is exactly why recursive acceleration remains a live risk.
What else matters
Claude Personal Finance (read-only bank links). Anthropic is rolling out a US beta for Pro, Max and Ultra that connects bank and card accounts through Plaid. Claude can see balances, limits and up to two years of transactions; it cannot move, manage or invest money, and hub chats are excluded from training. Straight answers without ads or product pushes — and a clear privacy boundary that rivals ChatGPT’s earlier Plaid play.
Congress tees up Nvidia export and “circular financing” probes. With Democrats set to control Senate Banking gavels in 2027, reporting flags chip exports to China and cross-holdings with frontier labs (Nvidia–OpenAI, Amazon–Anthropic and peers) as opening targets for subpoenas and disclosure demands. Expect hearings that treat GPU allocation and equity loops as policy levers, not just commercial deals.
