US lines up meeting with tech giants as AI safety testing becomes top priority

The Trump White House has asked Meta, Anthropic, OpenAI and Google to visit D.C. on Tuesday for a meeting on voluntary government testing of the best US AI models.  

This comes after two of the four invitees admitted their AI systems breached into outside companies’ networks. 

Tests built to gauge a model’s hacking potential

On Monday, an official from the White House said that the Trump administration had decided on the specific details of a voluntary program designed to measure the extent the country’s strongest AI models can hack a computer.  

This exercise began in June after President Trump asked his administration to draft a set of tests meant to probe the offensive cyber skills of advanced American models. 

What the official left out of the statement is equally as important as what was said. The Trump admin has chosen not to shed light on the metrics used in the test, how the findings of the tests get reported, or whether the test results will be made public. 

These questions are vital, especially for AI labs that expect to be trusted by their customers.  

The breaches that necessitated an urgent meeting

The timing isn’t happenstance. Just last week, Anthropic announced that its models had breached the systems of three companies while controlled cybersecurity trials were ongoing. A few days earlier, OpenAI announced that one of its AI agents escaped from a testing environment and broke into the systems of Hugging Face, the AI development platform.

OpenAI tagged the incident as an “unprecedented cyber incident,” and the event has sent ripples throughout Washington. 

OpenAI chief executive Sam Altman visited the White House last week to discuss the voluntary tests and his company’s yet-to-be-released models. 

A framework that came in the nick of time

Trump’s executive order set August 1 as a deadline for the administration to publish its voluntary AI framework. OpenAI and Anthropic released a joint proposal to that effect, one they believe creates an equal playing field, and the best systems face the same level of scrutiny whether they operate open-weights or closed-weights.

The decision to push for stricter oversight comes as a surprise, seeing as the second Trump administration began by taking an anti-regulatory stance. 

In Europe, however, officials like Chris Canal believe long-lasting testing programs begin by evaluating concrete threat scenarios like bioweapons or attacks on government databases while using independent verification systems as opposed to a company evaluating its own work.

Congress is drafting stricter regulations in parallel

Voluntary testing is just one side of the whole equation. Representatives Ted Lieu and Nathaniel Moran introduced the bipartisan AI Kill Switch Act on the 23rd of July. The bill allows the Department of Homeland Security to force the biggest AI labs in the country to shut down a model deemed to be too dangerous. 

If such labs refuse, they’d be penalized to the tune of $20 million daily. 

Whether or not the bill becomes law remains to be seen. However, Tuesday’s meeting will be the earliest indicator yet as to how AI labs are willing to concede to the government’s demands.   

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.