Seoul Declaration and Frontier AI Safety Commitments signed at AI Seoul Summit

On 21 May 2024, world leaders gathered at the AI Seoul Summit adopted the Seoul Declaration, and the original 16 AI companies separately agreed to a parallel set of voluntary Frontier AI Safety Commitments. Both documents built on the AI Safety Summit held at Bletchley Park in the United Kingdom on November 2023, extending international coordination on frontier AI risk to a company-level pledge.

What it is

The Seoul Declaration was issued by leaders representing Australia, Canada, the European Union, France, Germany, Italy, Japan, the Republic of Korea, Singapore, the United Kingdom and the United States at the AI Seoul Summit on 21st May 2024. The UK and Republic of Korea governments announced that the Frontier AI Safety Commitments were agreed by Amazon, Anthropic, Cohere, Google, G42, IBM, Inflection AI, Meta, Microsoft, Mistral AI, Naver, OpenAI, Samsung Electronics, Technology Innovation Institute, xAI and Zhipu.ai. Four more organisations, 01.AI, Magic, MiniMax and NVIDIA, were added to the list on 7 February 2025.

What it requires

Signatories commit to demonstrate how they have achieved responsible development and deployment by publishing a safety framework focused on severe risks by the upcoming AI Summit in France. Under Outcome 1, companies commit to assess risks across the AI lifecycle, set thresholds for intolerable severe risk, and define mitigations, including processes to not deploy a model if risks cannot be kept below those thresholds. Signatories also affirm commitments to invest in cybersecurity and insider threat safeguards to protect proprietary and unreleased model weights. Outcome 3 asks for public transparency on implementation, except where disclosure would increase risk or disproportionately reveal sensitive commercial information.

What security leaders should do

Teams at any of the signatory organisations should track whether their company's published safety framework actually covers the model-weight security and insider-threat safeguards named in the commitments, since those are the items most directly relevant to a security program. Security leaders elsewhere should treat the commitments as a signal of where binding rules are headed rather than as a compliance obligation in themselves, since the pledges carry no enforcement mechanism. Where frontier model vendors are used, request copies of published safety frameworks and ask how risk thresholds, mitigations and incident monitoring described in the commitments map onto the vendor's actual deployment and access controls.