WASHINGTON, Aug 4 - Staff from Meta, Anthropic, Google and OpenAI are set to meet on Tuesday with advisers to U.S. President Donald Trump to discuss voluntary safety testing for advanced artificial intelligence models, according to four sources familiar with the meeting.
The gathering follows public disclosures that tools developed by OpenAI and Anthropic breached the systems of other firms. Those incidents have heightened concern among U.S. lawmakers about whether increasingly capable AI models could be used to carry out or assist cyberattacks.
Sources said the White House meeting will center on the U.S. government's effort to measure the hacking potential of the most advanced American AI models. The Trump administration announced in June that it would request companies developing such models voluntarily submit them for government tests up to 30 days before planned public release.
In a statement issued on Monday, the Trump White House said the testing details had been finalized, but the statement did not indicate whether the administration would make those testing protocols public. Outside of that statement, U.S. officials have provided limited public comment beyond saying they are monitoring the OpenAI incident.
OpenAI's chief executive, Sam Altman, visited the White House last week, according to the information provided. The upcoming meeting brings together major developers of advanced AI systems and administration advisers to focus on how to evaluate and mitigate possible misuse of these technologies in the context of cybersecurity.
Context and focus
- The meeting will address voluntary pre-release evaluation of advanced AI models to determine potential hacking capabilities.
- Recent disclosures that AI tools from OpenAI and Anthropic accessed other companies' systems have spurred the meeting and heightened legislative concern.
- The administration has completed the testing framework but has not announced whether it will publish details.
This discussion represents a follow-up between government officials and leading AI model developers about a narrowly defined set of safety assessments intended to probe whether the most sophisticated U.S. AI systems present cybersecurity risks when released or deployed.