The White House will host AI companies Tuesday to discuss a framework for testing AI models. The companies will help evaluate the cybersecurity capabilities of their own systems. This is like asking a teenager to grade their own homework while you watch and nod.
Trump's executive order from June mandated this process. Officials spent two months developing criteria to measure whether advanced AI models pose security risks. The same companies that built these models will now participate in reviewing how those models should be tested. No conflict of interest there.
Imagine Ford showing up to a safety meeting at the DMV to help write the rules about whether Ford trucks catch fire too often. That's this meeting. Except the trucks can write code and the fire is a cyberattack that could shut down a power grid.
The framework will determine how officials assess AI capabilities that might threaten national security. The AI companies have a strong incentive to make sure the testing framework is rigorous and uncompromising. Just like how Philip Morris definitely wanted strict tobacco regulations.
This is the same administration that couldn't figure out how to use email without creating a congressional hearing. Now they're hosting the most sophisticated technology companies on Earth to collaborate on model-testing protocols. The people who thought you could nuke a hurricane will referee the entities that can generate synthetic bioweapons.
Retail traders will somehow find a way to lose money on this news. They'll buy calls on cybersecurity ETFs, miss the entire point of the meeting, and end up holding bags by Wednesday. The AI companies will leave with a testing framework that says "please check this box if your model will destroy humanity" and everyone will call it progress.
Tuesday's meeting won't change anything except the number of PowerPoint slides the government files under classified.
Photo by Anthony Roberts on Unsplash
Leave a Comment