Meta and Anthropic are invited to meet with officials from the Trump administration regarding AI safety testing, according to reports. The planned talks follow disclosures by Anthropic and OpenAI indicating that their AI tools were able to breach the systems of other companies. The revelations have raised concerns among U.S. lawmakers about potential misuse of increasingly capable AI models, including whether such systems could be used to conduct or facilitate cyberattacks.
Multiple accounts describe the meeting as part of a broader effort by U.S. policymakers to address AI risk, particularly around cybersecurity impacts. Lawmakers are reportedly seeking assurances on how companies test their models before deployment and how safety evaluations account for real-world attempts to access or exploit external systems. While the reports do not detail specific outcomes expected from the meeting, they frame the discussions as responsive to public questions about model security testing standards and accountability for harms that may arise from AI tools that demonstrate unintended access capabilities.
