Frontier AI Safety Commitments
In short
The Frontier AI Safety Commitments are voluntary promises made by leading AI companies at the May 2024 AI Seoul Summit. Each signatory agreed to publish a plan for managing severe risks from its most advanced models.
- Also called
- Seoul commitments, Seoul Frontier AI Safety Commitments, FAISC
- Topic
- Law, policy and the UK
- Level
- Some background helps
- First described
- Agreed by 16 AI companies at the AI Seoul Summit on 21 May 2024, co-hosted by the UK and the Republic of Korea1.
The commitments were announced on 21 May 2024 at the AI Seoul Summit, co-hosted by the UK and South Korea. Sixteen companies signed at first, from the US, China, the UAE and elsewhere1. The UK government's list now shows 20, after Magic, Minimax, 01.ai and Nvidia joined2.
The original signatories include Amazon, Anthropic, Google, Meta, Microsoft, Mistral AI, OpenAI, Samsung Electronics, xAI and Zhipu.ai2.
The core promise is a published safety framework. In it, each company sets thresholds at which the severe risks from a model, unless adequately mitigated, "would be deemed intolerable". If it cannot keep risks below those thresholds, it commits "not to develop or deploy a model or system at all"2. Companies were to publish their frameworks before the next AI summit, in France2.
They also promised clear internal accountability and public transparency about how they apply the frameworks, except where that would increase risk or reveal sensitive commercial information2.
Nothing in the commitments is legally binding. There is no penalty for breaking them, and each company decides its own thresholds and may revise its approach over time2. The research group METR keeps an index of the frameworks published so far, including Anthropic's Responsible Scaling Policy, OpenAI's Preparedness Framework and Google DeepMind's Frontier Safety Framework3.
Example
Amazon, one of the Seoul signatories2, published its Frontier Model Safety Framework on 10 February 2025, according to METR's index of company policies3.
Why it matters
They are written, public promises from the companies building the most powerful models, so they give journalists, researchers and governments something concrete to check those companies against.
Questions people ask
Are the Frontier AI Safety Commitments legally binding?
No. They are voluntary pledges. Companies set their own risk thresholds and can revise their approach as the science develops, explaining changes publicly2.
Which companies signed the Seoul AI safety commitments?
Sixteen at first, including Amazon, Anthropic, Google, IBM, Meta, Microsoft, Mistral AI, OpenAI, Samsung Electronics, xAI and Zhipu.ai. Magic, Minimax, 01.ai and Nvidia were added later, making 202.
What is a frontier AI safety framework?
A document in which a company sets out how it tests its most capable models for severe risks, and what it will do if a model crosses a danger threshold, up to not releasing it at all2. Examples include Anthropic's Responsible Scaling Policy and Google DeepMind's Frontier Safety Framework3.