At the AI Seoul Summit, announced by the UK and the Republic of Korea on 21 May 2024, sixteen leading AI companies agreed to a set of voluntary Frontier AI Safety Commitments. The original signatories were Amazon, Anthropic, Cohere, Google, G42, IBM, Inflection AI, Meta, Microsoft, Mistral AI, Naver, OpenAI, Samsung Electronics, the Technology Innovation Institute, xAI, and Zhipu.ai, with Magic, MiniMax, 01.AI, and NVIDIA joining later.
The commitments ask each company to publish a safety framework describing how it will manage risks from its frontier models. Companies pledged to assess the risks their frontier models pose across the AI lifecycle and to “set out thresholds at which severe risks posed by a model or system, unless adequately mitigated, would be deemed intolerable.” Crucially, they agreed not to “develop or deploy a model or system at all” if mitigations cannot keep risks below those thresholds. The commitments also covered red-teaming, cybersecurity protection of model weights, mechanisms to disclose AI-generated content, and sharing information with trusted actors including governments.
These pledges built on the Bletchley Declaration from late 2023 and were the first time a large group of frontier developers publicly committed to defining “intolerable” risk levels and to halting development when they cannot be managed. For businesses and the public, the commitments matter because they shifted some of the responsibility for setting safety limits onto the companies themselves, while leaving open the question of how compliance would be verified.