As I’ve written before, one reason there’s so much confusion in AI news right now is that it’s unclear what AI companies are allowed to do and what they aren’t allowed to do. But there is reportedly a voluntary agreement in the works with Big AI that could smooth things over considerably (your mileage may vary on whether this is a good thing or not).
According to the Financial Times, “as early as next week” the Trump administration and several major US frontier AI companies are expected to announce a set of standards for frontier AI models, particularly with regard to cybersecurity capabilities. The report cites “people familiar with the conversations”—unnamed leakers, in other words.
One of the FT’s unnamed sources said the Center for AI Standards and Innovation (CAISI), which is under the Commerce Department, and the National Security Agency (NSA), which is under the Pentagon, will formally be at the center of these standards.
On June 12, the US gave Anthropic an export control directive, which essentially shut down its latest publicly released model, and kept it offline for the rest of June. OpenAI, apparently concerned that something similar could happen and spoil its plans, has paused the release of its latest models as a precaution.
At the beginning of the Trump 2.0 administration, Vice President J.D. Vance signaled a laissez-faire approach to AI regulation. This has changed significantly now, with the White House actions against Anthropic, its executive order regarding AI, and now these standards, which appear to be a formalization of some aspects of the order.
According to the order, the government is expected to:
“…develop and maintain a classified benchmarking process to assess the advanced cyber capabilities of AI models and determine the threshold at which AI models should be designated ‘covered frontier models’ for the purposes of this order, sharing such assessments with AI developers and researchers as appropriate.”
If the benchmarking process is indeed classified, that means the public will not know what standards Big AI is being held to. However, shared practices regarding security measures across many companies will make it easier to achieve at least part of the standards that have been agreed upon.
It is not entirely clear which companies will be parties to this voluntary investigation agreement. The FT article mentions Anthropic, OpenAI, Amazon, Microsoft and Google. Interestingly, it doesn’t mention Meta, and about a week ago, other unnamed sources familiar with these negotiations reportedly leaked that Meta was a holdout, and the Trump administration was working overtime to buy Meta.
<a href