The administration hosts a meeting with Anthropic, OpenAI, and Google to review a new cybersecurity framework for advanced AI models.
The administration will host a new meeting on Tuesday with representatives from Anthropic, OpenAI, and Google to review a finalized framework for assessing the cybersecurity capabilities of advanced artificial intelligence models. This voluntary program, ordered by President Trump in June, establishes a process for developers to determine if their models qualify as "covered frontier models." Under the framework, participating developers can provide the government with early access to their models for up to 30 days before public release. This allows the government and technology companies to evaluate how powerful models might identify software vulnerabilities or execute cyberattacks. The administration announced that the Treasury Department, National Security Agency, and Cybersecurity and Infrastructure Security Agency will establish a classified benchmarking process to assess these capabilities. The framework is designed to be voluntary, ensuring that the administration does not establish a mandatory federal licensing or preclearance requirement for developers. This initiative comes as AI systems become increasingly autonomous and demonstrate the ability to identify and exploit vulnerabilities independently.