The weekend’s call to slow frontier AI development is already moving equities. At 4:46 a.m. ET, Nasdaq-100 futures were down 1.72%; software names moved the other way, with ServiceNow up 3% and Adobe and Workday 2.5%. The cross-section is consistent with investors reducing the expected pace of AI spending and disruption, rather than selling technology indiscriminately. It is an early premarket observation, with oil and rates also weighing on sentiment. Reuters, this morning
Dario Amodei’s proposal contains a commitment worth reading closely: embedded external evaluators with employee-like access to Anthropic’s workspaces and training processes, and the right to publish without its editorial control. They would examine how models are made as well as the finished models. Anthropic says it intends to invite a team soon; the arrangement is not yet an operating audit. Amodei’s essay
There would be exceptions for security, legal privilege and confidential information. Evaluators could publicly state that a redaction removed something important to their conclusions. That last provision gives the promise some substance: the company could restrict disclosure without necessarily controlling the account of the restriction.
Established laboratories benefit if safety requirements raise competitors’ costs; independent scrutiny can also constrain the incumbents. Much will turn on who is appointed, what access they receive and how the first serious disagreement is handled.
His broader proposal requires industry coordination and government help with antitrust constraints. That is considerably further from implementation than the unilateral evaluator commitment. Full proposal and exceptions