Summary
On July 21, 2023, President Joe Biden announced that seven companies, Amazon, Anthropic, Google, Inflection, Meta, Microsoft and OpenAI, had agreed to a set of voluntary commitments on the safety, security and transparency of their AI systems. The companies committed to security testing by internal and independent experts before release, to protecting unreleased model weights, to sharing risk information with each other and with governments, and to developing watermarking for AI-generated content. The commitments were not legally binding, and the White House presented them as a step toward an executive order and legislation.
What Happened
Biden convened the seven companies at the White House and spoke about the commitments, which he called "real and concrete." The White House fact sheet said the commitments, "which the companies have chosen to undertake immediately," rested on three principles: safety, security and trust.
Under safety, the companies committed to internal and external security testing of their AI systems before release, carried out in part by independent experts and covering risks such as biosecurity and cybersecurity, and to sharing information on managing AI risks across the industry and with governments, civil society and academia. Under security, they committed to invest in cybersecurity and insider-threat safeguards to protect "proprietary and unreleased model weights," and to make it easier for third parties to find and report vulnerabilities. Under trust, they committed to develop technical mechanisms, such as watermarking, to let users know when content was AI-generated; to report publicly on their systems' capabilities, limitations and appropriate uses; to prioritize research on societal risks such as bias and privacy; and to deploy advanced AI to help address problems such as cancer and climate change.
The fact sheet said the administration was developing an executive order and would pursue bipartisan legislation, and that it had consulted on the commitments with 20 other countries. It placed the announcement after a May meeting at which Biden and Vice President Kamala Harris had met the chief executives of Google, Anthropic, Microsoft and OpenAI. The Guardian reported that the commitments were not legally binding and quoted critics, including Common Sense Media founder James Steyer, who said that "many tech companies do not actually walk the walk on a voluntary pledge to act responsibly."
Why It Matters
The commitments put specific safety practices in a document that the US government and the leading frontier developers both endorsed, and they defined the vocabulary of voluntary frontier governance that followed: pre-release red-teaming, protection of model weights, watermarking and public model reports. Four of the seven companies founded the Frontier Model Forum five days later, Anthropic published its Responsible Scaling Policy two months later, and OpenAI and Google DeepMind followed with comparable frameworks that tied model releases to safety evaluations.
The commitments also showed the limits of that approach. They had no enforcement mechanism, and each company decided for itself how to meet them. The executive order the fact sheet promised arrived on October 30, 2023, and turned some of the testing and reporting elements into requirements for the most capable models.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge filed and not resolved by a correction. Corrected = a challenge was upheld and the entry corrected; see its revision history. Superseded = replaced by a later entry. Unresolved = dispute still open.