World’s Leading AI Companies Make Historic AI Safety Pledge
In a move to ensure the safe and responsible development of advanced AI, the governments of the United Kingdom and the Republic of Korea announced today that 16 leading AI organizations have agreed to the Frontier AI Safety Commitments. This significant initiative marks a collective effort to address the potential risks associated with frontier AI technologies, building on the foundational Bletchley agreements.
The organizations that have endorsed these commitments include:
– Amazon
– Anthropic
– Cohere
– Google
– G42
– IBM
– Inflection AI
– Meta
– Microsoft
– Mistral AI
– Naver
– OpenAI
– Samsung Electronics
– Technology Innovation Institute
– xAI
– Zhipu.ai
Core Commitments for Safe AI Development
These prominent AI entities have pledged to develop and deploy their frontier AI models and systems responsibly, adhering to a framework designed to manage severe risks. This commitment involves several key practices and goals:
1. Risk Identification and Management: Organisations will conduct thorough risk assessments throughout the AI lifecycle, including during both the training and deployment phases. These assessments will consider model capabilities, deployment contexts, and mitigation effectiveness against potential misuse. Independent evaluations will also be utilized where appropriate.
2. Defining Risk Thresholds: Clear thresholds for severe risks will be established, with input from trusted entities and in alignment with international agreements. If a model or system approaches or exceeds these thresholds, organisations will implement necessary mitigations or, if risks cannot be adequately managed, cease development or deployment.
3. Investment in Risk Mitigation: Continuous investment will be made in advancing risk identification, assessment, and mitigation techniques. Organizations will monitor and update their risk management strategies to align with emerging best practices and international standards.
4. Transparency and Accountability: Organizations will provide public updates on their safety frameworks and how they achieve their commitments, ensuring transparency in their processes. Detailed, sensitive information will be shared with trusted entities, including home governments, where public disclosure is not feasible.
5. External Engagement: The involvement of external actors, such as governments, civil society, and academics, will be integrated into the risk assessment process to enhance the safety and accountability of frontier AI models and systems.
Key Outcomes and Implementation Strategies
The commitments focus on achieving three primary outcomes:
– Effective Risk Management: Organizations will implement robust processes to identify, assess, and manage risks associated with frontier AI models. This includes setting and monitoring risk thresholds, implementing mitigations, and ceasing operations if necessary to keep risks within acceptable bounds.
– Accountability in AI Development: Adhering to the outlined commitments, organizations will develop internal governance frameworks to ensure ongoing compliance and accountability in AI development and deployment.
– Transparency to External Actors: Organizations will maintain transparency in their safety practices, sharing implementation details publicly and with trusted entities to foster trust and collaboration in managing AI risks.
Looking Ahead
As part of the commitments, the 16 companies will publish their safety frameworks ahead of the AI Action Summit in France in early 2025. These frameworks will outline risk thresholds and mitigation strategies, providing a transparent roadmap for the safe development and deployment of frontier AI models.
