On Sept. 29, 2026, President Donald Trump and executives representing Google, Meta, OpenAI, Anthropic, Nvidia and xAI presented a voluntary AI safety pact at the White House. The six companies were reported to have signed it with the U.S. government. Its reported framework includes internal safeguards, outside evaluations and independent oversight.
Four reported layers of oversight
The pact’s described safeguards range from controls inside each company to outside evaluations and an independent committee associated with company boards.
| Safeguard | Who is involved | Described role |
| Company security controls | The companies | Protect technical systems against unauthorized access and help keep AI systems aligned with developers’ instructions. |
| Internal risk-monitoring teams | Teams within the companies | Monitor risks associated with AI systems. |
| External evaluations | Independent outside evaluators | Assess risks, including cybersecurity and biosecurity concerns. |
| Independent committee | A committee associated with company boards | Provide an additional layer of oversight. |
The stated goals include preventing unauthorized access to technical systems and keeping AI aligned with developers’ instructions. The risk areas include cybersecurity and biosecurity.
What voluntary means for the agreement
The pact is voluntary, not a federal law, and carries no fines or legally binding duty to comply. It sets out a shared commitment rather than a statutory requirement for the participating companies.
