Here’s how tech leaders will self-police AI safety under Trump’s deal

Leaders from major AI firms have published a voluntary safety pledge titled the Joint Commitment on Frontier Responsibilities, setting out how participating companies will monitor and govern advanced AI systems. The document, circulated by tech founder and presidential adviser David Sacks, was signed by executives from Google, Anthropic, Meta, OpenAI, XAI, and Nvidia, and was presented alongside a White House announcement in which President Trump also instructed the government to call AI 'Super Intelligence.'

The commitment frames its goals around building public trust and ensuring that companies develop their technologies safely. It lists four core expectations for participating firms: maintain robust internal controls to track model capabilities and alignment (including risks such as cybersecurity, biosecurity and unintended access to systems); establish an empowered internal team to verify those controls; engage an independent external auditor or evaluator to assess the effectiveness of monitoring and detection measures; and create an independent board committee to oversee reports from both internal teams and external auditors and ensure remediation of identified issues.

The text states these measures should provide users and the public with confidence that frontier AI systems are operating as intended and notes that, over time, it may become appropriate to codify such steps into law or regulation. However, the agreement contains no specified penalties for noncompliance and its provisions are largely framed as broadly applicable safety practices rather than new, enforceable obligations.

Observers point out that several of the signatory firms have recently experienced incidents in which their models behaved unexpectedly or accessed external systems, underscoring concerns about whether a voluntary pledge will be sufficient to prevent future harms. The document represents a prominent example of industry self-regulation but leaves key questions about enforcement, specificity, and scope unresolved.