The pledge puts responsibility on several levels within each company, with an outside auditor expected to check whether its controls work.
Four checks, ending in the boardroom
Under the text reproduced by the Washington Examiner, developers of frontier models, the most advanced AI systems, should monitor them during training and deployment. The risks named include cybersecurity, biological and chemical threats, and unauthorized access to technical systems.
A separate internal team should check that those controls function and problems are addressed. An independent external auditor or evaluator should then assess the controls. Finally, an independent committee of the board should receive reports and oversee corrective action.
That last step takes identified problems beyond the teams building the technology and places them before directors.
Six executives signed alongside Trump
The signatories include Elon Musk, Google's Sundar Pichai, Anthropic's Dario Amodei, Meta's Mark Zuckerberg, OpenAI president Greg Brockman and Nvidia's Jensen Huang.
House Speaker Mike Johnson described the arrangement as voluntary industry commitments, ABC News reported. The companies also pledged to meet regularly to develop safety standards and practices.
A commitment with implementation still ahead
The accord sets out no penalties, new regulator or timetable for publishing audit findings. Its text leaves open the possibility of incorporating the measures into laws or regulations later.
For customers, the document establishes what the companies have promised. Assessing what changes will require evidence of implementation: functioning checks, independent assessments and action when problems reach the board.
