Google, Anthropic, Meta, OpenAI, xAI and Nvidia signed a joint White House safety pledge on September 29. The document published by the official Rapid Response account bears Donald Trump's signature alongside Sundar Pichai, Dario Amodei, Mark Zuckerberg, Greg Brockman, Elon Musk and Jensen Huang.
The companies describe four layers: controls during training and deployment, an empowered internal team to check them and remedy failures, independent external evaluators, and an independent board committee. The controls are intended to detect dangerous capabilities and prevent unintended hacking or resource access, among other risks.
The participants plan regular meetings on standards and best practices. The document anticipates possible future laws or regulations but makes the commitment independently of them. It is voluntary; this record does not establish published audit results or completed implementation.
Alignment researcher Victoria Krakovna called some commitments promising in her own post and reproduced the text. Her approval is not an employer position or evidence of another agreement. AI Incidents records the accord as a risk signal, not a new AI incident or proof that the risks are under control.