US tech giants invited to discuss AI security tests at White House

Technology giants Meta, Anthropic, Google, and OpenAI will convene at the White House on Tuesday to deliberate a voluntary framework for AI security testing. According to Reuters, the Trump administration has finalized the specifics of the framework, aimed at granting government access to leading AI models for hacking capability evaluation.

The Information previously confirmed the White House's Tuesday meeting with the AI industry, a shift from the administration's non-interventionist stance. The details of this framework, potentially derived from the Trump administration's June AI executive order, have not been publicly disclosed. Companies agreeing to the framework might allow federal testing of their "covered frontier models" 30 days prior to granting access to selected trusted partners.

The framework aims to identify models with significant cyber capabilities and assess their safe deployment to bolster critical infrastructure. However, it remains voluntary and does not grant the government the power to license models or inhibit their public release. The administration's attempt to introduce regulatory safeguards on AI systems aims to balance innovation with security, reflecting concerns about China's competitive edge in AI.

The framework, directed to be developed by August 1, prioritizes cybersecurity across federal agencies and establishes a cybersecurity clearing house to identify and resolve software vulnerabilities.