On Tuesday, President Trump signed an executive order directing the federal government to vet the national security risks of advanced AI models before public release. AI companies are asked, voluntarily, to share frontier models with the government up to 30 days before broader release. The order also directs agencies to develop benchmarks for assessing AI models' cyber capabilities and to stand up an AI cybersecurity clearinghouse, led by the NSA.

Yesterday, OpenAI published a policy paper calling for a mandatory version of that framework, but with a critical difference. Where the White House order places oversight with the NSA, OpenAI wants the responsibility with CAISI, a civilian office inside NIST at Commerce. OpenAI and Anthropic already share model information with CAISI. Neither has a relationship with the NSA. Sam Altman was in Washington Wednesday pushing that case directly with White House officials and lawmakers. OpenAI's Chris Lehane was clear on where the company stands: mandatory evaluations, yes. Intelligence community oversight with classified benchmarks, no.

Worth noting: last week I wrote about how the AI lab CEOs called for this kind of binding oversight framework under oath in 2023 — then quietly walked it back by 2025. OpenAI is back asking for mandatory evaluation requirements. The ask has come full circle.

Also, at a Senate Finance Committee hearing on Wednesday, Sen. Mark Warner pressed Treasury Secretary Scott Bessent on the voluntary nature of the clearinghouse, warning it would put national security at risk. Bessent told Warner the only change from prior drafts was a tighter deadline. Warner pushed back, saying the draft he had seen was "stronger than strictly voluntary." The final order explicitly states nothing in it should be construed as mandatory.

The timing matters beyond the politics. The same week Washington debates who should oversee AI before deployment, federal agencies are being called out for having no framework to assess what AI is doing after deployment. Today’s Intelligence Report highlights how the government can capture AI outcomes informing impact and accountability.

Let's get into it.