OpenAI Pushes for Greater Role for CAISI in AI Evaluation

OpenAI’s new proposal comes as its CEO Sam Altman descends on Washington for a series of Wednesday meetings with White House officials and key lawmakers in both political parties.

The White House did not immediately respond to a request for comment on OpenAI’s new plan.

In a conversation with reporters Wednesday morning, top OpenAI executive Chris Lehane praised the general direction of Trump’s new executive order, calling it a “validation” of efforts by his company to develop a regulatory framework for advanced AI. But Lehane also suggested that he, Altman and others within the company will push the Trump administration and Congress to give CAISI a greater role in the AI evaluation process.

“You also have CAISI that has developed into something and has the capability to really do the type of sophisticated testing,” Lehane said. He noted that OpenAI, Anthropic and other top companies have already agreed to share information about new AI models with CAISI. “They’ve sort of developed that relationship with these companies, and so that already exists,” Lehane added. The NSA currently has no such relationship with top AI companies.

The OpenAI executive also indicated his company has concerns with the new White House plan to develop and maintain a “benchmarking” process for advanced AI models on a classified basis — a provision in the order that could make it difficult for companies to determine when and whether their new models will come under scrutiny from the NSA and other intelligence agencies.

“I think one of the items here is when do you hit the capability threshold?” Lehane asked. “I think that’s a big part of what the conversation will be — can you establish some criteria of what that is?”

Lehane said he expects Altman is “certainly going to talk about our ideas and thinking on this” during meetings with the White House on Wednesday. He noted that Trump’s new order gives agencies 60 days to work out details — suggesting OpenAI has time to shape the final outcome — and that key lawmakers on Capitol Hill are also weighing plans to empower and expand CAISI’s ability to conduct AI evaluations.

The OpenAI executive added that the company plans to push policymakers in Washington to create mandatory evaluation processes for AI companies that develop advanced systems. “We don’t think any specific lab should be making that decision unilaterally,” Lehane said.

But he ultimately shied away from other mandatory regulations — including a potential requirement for AI companies to get a green light from the government before they release new models — contending that it’s too early to discuss more aggressive regulations until a robust evaluation framework is put in place.

“I think you have to do these first steps before you can start to figure out the additional pieces,” Lehane said.