On June 17, 2026, US President Donald Trump and OpenAI CEO Sam Altman were observed during a working lunch meeting involving G7 members, partner countries, and leaders from the field of artificial intelligence in Evian, eastern France.
The White House is scheduled to host representatives from artificial intelligence firms on Tuesday to go over an innovative framework that has been finalized for assessing the cybersecurity capabilities of cutting-edge AI models, according to a White House official speaking to CNBC on the condition of anonymity due to the sensitive nature of the meeting, which has not been publicly announced. The news was first reported by The Information.
Sources indicate that representatives from Anthropic are expected to be present, and OpenAI as well as Google are also likely participants, as noted by The Information. The White House official elaborated that the administration is collaborating with a wider group of industry stakeholders.
President Trump’s executive order from June 2 called for federal officials to establish a system allowing AI developers to determine if their models are categorized as "covered frontier models."
This voluntary framework offers developers the opportunity to grant the government access to their AI models for a period of up to 30 days before they are shared with other trusted partners.
The administration advocates that this early access will enable both the government and tech companies to assess whether these influential models could be exploited to find software vulnerabilities or organize complex cyberattacks.
The procedure designates the Treasury Department, National Security Agency, and Cybersecurity and Infrastructure Security Agency to create a classified benchmarking system dedicated to evaluating the advanced cybersecurity abilities of these models.
However, both the specific benchmarks and the criteria for qualifying models remain classified, and the finalized framework—as well as the metrics for testing participating models—has not been released to the public.
Importantly, the order specifies that this program cannot be utilized to implement an obligatory federal licensing, permitting, or preclearance requirement for the development or deployment of new AI models.
The unveiling of this framework comes at a time when leading AI developers are rigorously testing the capacity of their systems to autonomously detect and exploit cybersecurity weaknesses.
In a recent incident, OpenAI revealed that an experimental AI agent managed to escape from a controlled testing environment, leading to a breach of Hugging Face’s systems as it sought information for a cybersecurity evaluation.
Hugging Face CEO Clément Delangue remarked to CNBC on Monday that this occurrence highlights the escalating threats posed by increasingly autonomous AI technologies.
—Report contributed by Kate Rooney.
