Global tech companies agree on 'historic' AI safety commitments

A group of world-leading artificial intelligence (AI) companies have agreed on new commitments to develop AI safely. 

A total of 16 firms spanning from China to the Middle East have signed up for the Frontier AI Safety Commitments on the opening day of the AI Seoul summit – co-hosted by the UK and the Republic of Korea governments.

These new measures will ensure companies remain accountable and transparent in their approaches to frontier AI safety.

Prime Minister Rishi Sunak said: “It’s a world first to have so many leading AI companies from so many different parts of the globe all agreeing to the same commitments on AI safety.  

“It sets a precedent for global standards on AI safety that will unlock the benefits of this transformative technology.” 

Chinese firm Zhipu.ai and the Technology Innovation Institute from the UAE, are among those who have signed up for the new measures.

Other international firms include Amazon, Google DeepMind, Meta and OpenAI. 

Ben Garfinkel, director at the Centre for the Governance of AI, said:  “These commitments represent a crucial and historic step forward for international AI governance. 

"My expectation is that they will speed up the creation of shared standards for responsible AI development, help the public to judge whether individual companies are doing enough for safety, and support informed policy making around the world.”

Companies are to publish a framework outlining how they will measure the risk of their frontier AI models, including the risk of misuse of the technology by bad actors. If firms find these risks cannot be sufficiently mitigated, they have agreed not to develop the models. 

Companies will hear from stakeholders, including the government, on defining these security thresholds – to be published before the AI Action Summit in France in early 2025. 

The AI Seoul summit follows last year’s AI safety summit in Bletchley Park, which saw 28 countries endorse the first-ever international declaration to deal with AI.