Updates from the AI Safety Summit and the Way Ahead

On the 1st and 2nd of November 2023, the United Kingdom of Great Britain and Northern Ireland chaired the inaugural edition of the AI Safety Summit (“AI Summit”) at the Bletchley Park, Buckinghamshire. The venue of the AI Summit was symbolic, with Alan Turing cracking the Enigma cipher and the world’s first electronic computer being at used at Bletchley Park[1]

The AI Summit was organised with the intent to assess the risks posed by the transformational technology of Artificial Intelligence (AI) by way of a global endeavour.

Scope of the AI Summit[2]

The AI Summit focused on the potential risks arising from two types of AI:

  1. Frontier AI, which is a highly capable generation of general-purpose AI models that can perform a wide variety of tasks and match or exceed the capabilities present in today’s most advanced AI models. These would include cutting-edge Large Language Models (LLMs)[3].
  2. Specific narrow AI which can hold potentially dangerous capabilities, for instance, AI models designed for bioengineering.

Types of risks addressed through the AI Summit

The AI Summit sought to address two categories of risks:

  1. Misuse risks. For example, a situation where a bad actor comes to be aided by new AI capabilities in biological or cyber-attacks, development of dangerous technologies, or critical system interference. Unchecked, this could create significant harm, including the loss of life.
  2. Loss of control risks that could emerge from advanced systems that we would seek to be aligned with our values and intentions.

Objectives of the AI Summit

The AI Summit was conceptualised with five objectives:

  1. A shared understanding of the risks posed by frontier AI and the need for action.
  2. A forward process for international collaboration on frontier AI safety, including how best to support national and international frameworks.
  3. Appropriate measures which individual organisations should take to increase frontier AI safety.
  4. Areas for potential collaboration on AI safety research, including evaluating model capabilities and the development of new standards to support governance.
  5. Showcase how ensuring the safe development of AI will enable AI to be used for good globally.

A Snapshot of the AI Summit[4]

  • In order to further objective 1 of objective 2 of the AI Summit, the States participating in the AI Summit agreed on the Bletchley Declaration, which is in the nature of an initial mutual understanding on the need for global collaboration in combating the risks posed by AI, on frontier AI and the risks associated with it, and crystalised the commitment by countries to work in an inclusive manner to ensure human-centric, trustworthy and responsible AI that is safe.
  • The Bletchley Declaration also commits countries to further collaborate on establishing a shared scientific and evidence-based understanding of the relevant risks. As the first initiative towards this, the UK shall deliver a “State of the Science Report” on frontier AI[5]. The UK’s Department for Science, Innovation, and Technology has commissioned Yoshua Bengio, a Turing Award-winning AI scientist and member of the United Nations’ (UN) Scientific Advisory Body, to Chair the Report’s writing group, supported by a diverse group of leading AI academics, advised by an inclusive, international Expert Advisory Panel, with representatives from participating countries.
  • Discussions regarding objective 3 centered on the involvement of different actors in developing Frontier AI and addressing the impacts of the AI. The discussion highlighted the importance of collaboration and exchange of information between these different actors. The Participants discussed the publication of the AI safety policies by leading frontier AI developers pondering ways to establish trust through the development and disclosure of such policies.
  • In order to further objective 4, the states concurred that the government and frontier AI developers should align their responsibilities and jointly strategise for safety testing. UK has taken a notable step by inaugurating the world’s first AI Safety Institute, aiming to assess potential risks associated with AI. The initiative invites companies to engage in collaboration and participate in pre-deployment testing of their frontier AI models alongside the institute. The players recognised the need to collaborate on testing the next generation AI models against a range of critical national security, safety and societal risks and made a joint statement on safety testing.[6] Participants explored more ambitious policies, including legally binding existing voluntary commitments, establishing international safety standards for safety, implementing pre-deployment safety testing with defined gates, involving governments in early model testing, setting standards for failure rates, ensuring an off-switch for critical AI, addressing safety in open-source models, mitigating election interference risks, and emphasizing AI equity.
  • To advance objective 5, participants expressed a collective commitment to leverage the transformative power of frontier AI for societal and economic benefit. They emphasized inclusive design and discussed AI’s positive impact on sectors like healthcare and education. Participants advocated for making AI inclusive and accessible and welcomed international collaboration.

Way Forward:[7]

The states identified the following themes as proactive measures for the establishment of a secure frontier in AI.

  1. The need to build a shared understanding of frontier AI, addressing challenges of fragmentation and incomplete understanding of frontier AI. Participants emphasized government leadership in fostering public trust through a clear understanding of the technology. They committed to collaborating on the State of the Science Report, aiming to enhance understanding of Frontier AI beyond Summit attendees, and aligning with initiatives like the United Nations AI Advisory Body, Global Partnership on Artificial Intelligence (“GPAI”), and the Organisation for Economic Co-operation and Development (“OECD”).
  2. The need for an inclusive approach to address frontier AI and associated risks, emphasizes equitable development of AI bridging the digital and development divide, multi-stakeholder collaboration, and engagement with diverse initiatives. This includes tackling challenges faced by developing countries, women and other minority groups and promoting ‘AI for Good’ to achieve the UN Sustainable Development Goals.
  3. The importance of addressing current AI risks alongside those at the frontier, acknowledging the immediate threats such as spreading false narratives, electoral interference, crime, inequality, amplifying biases and discrimination. International initiatives, including the Council of Europe’s treaty on AI, G7 commitments, and the Responsible AI in the Military Domain Summit, underscored the global nature of these challenges.
  4. The value of standardization and interoperability in AI was discussed, with participants emphasizing the benefits of shared frameworks. They encouraged the development of interoperable approaches to mitigate frontier AI risks and promote inclusive AI benefits. Participants highlighted the role of multistakeholder organizations like the OECD and GPAI, in providing guidance for better interoperability of AI application, development and governance. Common principles and codes, such as those in the OECD Recommendation on AI and G7 Hiroshima AI Process, were seen as essential for establishing a baseline and fostering further multistakeholder engagement in AI development.
  5. The need to develop the broader AI ecosystem, including skills, talent, and physical infrastructure, was emphasized by participants. Key priorities include ensuring access to skills for designing, deploying, and using AI, along with discussions on infrastructure, data access, and environmental sustainability. Participants highlighted the risks of market concentration and the importance of sharing best practices.

Acknowledging the importance of ongoing global cooperation, participants pledged to reconvene in the future. They also appreciated the announcement that the Republic of Korea will co-host a virtual summit on AI in the next six months, followed by an in-person summit hosted by France a year later.



[1] Prime Minister Rishi Sunak’s statement from the AI Safety Summit at Bletchley Park- https://www.gov.uk/government/speeches/prime-ministers-speech-at-the-ai-safety-summit-2-november-2023

[2] Introduction to the AI Safety Summit- https://assets.publishing.service.gov.uk/media/65255655244f8e00138e7362/introduction_to_the_ai_safety_summit.pdf

[3] A type of AI algorithm that can understand, summarize, generate, and predict new content.

[4] Chair’s Summary of the AI Safety Summit 2023, Bletchley Park, https://assets.publishing.service.gov.uk/media/6543e0b61f1a60000d360d2b/aiss-chair-statement.pdf

[5] “State Of The Science” Report to understand capabilities and risks of Frontier AI, https://assets.publishing.service.gov.uk/media/6543b759d36c910012935cad/aiss-statement-state-of-science-report.pdf

[6] Safety Testing- Statement of session outcomes- https://assets.publishing.service.gov.uk/media/6544ec4259b9f5001385a220/aiss-statement-on-safety-testing-outcomes.pdf

[7] Chair’s Summary of the AI Safety Summit 2023, Bletchley Park, https://assets.publishing.service.gov.uk/media/6543e0b61f1a60000d360d2b/aiss-chair-statement.pdf.