White House to host AI companies to review new model-testing framework

The White House announced it will convene leading artificial‑intelligence firms to review a voluntary framework aimed at testing the cybersecurity resilience of next‑generation AI models.

Purpose of the meeting

The gathering is intended to gather industry input on a set of best‑practice guidelines that could help identify vulnerabilities in large language models and other advanced AI systems before they are deployed at scale.

Key participants

Representatives from major AI developers, including OpenAI, Anthropic, Google DeepMind, and Microsoft, are expected to attend, alongside officials from the Office of Science and Technology Policy and the Cybersecurity and Infrastructure Security Agency.

Framework highlights

The draft framework proposes a structured testing process that covers threat modeling, adversarial prompt injection, data poisoning, and model‑output leakage. Participation would be voluntary, but companies that adopt the guidelines could receive a “secure AI” designation from the administration.

• Define threat scenarios specific to AI outputs
• Establish standardized penetration‑testing protocols for model APIs
• Create a reporting mechanism for discovered vulnerabilities
• Encourage transparent disclosure of security findings

Potential impact

If widely adopted, the framework could set a benchmark for AI safety that other nations might emulate, fostering a more secure global AI ecosystem while mitigating risks of malicious exploitation.

“Proactive testing is essential to stay ahead of adversaries targeting AI systems,” said a senior White House official.

The administration plans to release a revised version of the framework later this year, incorporating feedback from the upcoming meeting and ongoing consultations with cybersecurity experts.

CNBC coverage of White House AI companies voluntary framework meeting