We’re strengthening the Frontier Safety Framework (FSF) to help identify and mitigate severe risks from advanced AI models.
Google DeepMind recently published an article titled "Strengthening our Frontier Safety Framework," detailing the third iteration of their Frontier Safety Framework (FSF). This update introduces a Critical Capability Level (CCL) focused on harmful manipulation, addressing AI models with potent manipulative capabilities that could be misused to systematically alter beliefs and behaviors in high-stakes contexts. Additionally, the FSF now includes protocols for machine learning research and development CCLs, emphasizing models that could accelerate AI research to destabilizing levels. The framework also refines its risk assessment process, providing more detailed evaluations to identify and mitigate severe risks from advanced AI models. (deepmind.google)
In parallel, the Frontier Model Forum (FMF) has been actively developing and refining frontier AI frameworks to manage potential severe risks associated with advanced AI systems. Their technical report series, including "Frontier Mitigations" and "Risk Taxonomy and Thresholds for Frontier AI Frameworks," outlines methodologies for identifying, managing, and mitigating large-scale risks to public safety and national security stemming from frontier AI development and deployment. (frontiermodelforum.org)
These initiatives reflect a growing commitment within the AI community to establish robust safety frameworks that proactively address the challenges posed by advanced AI technologies.
Discover DeepMind, a world-leading AI research lab by Google. Learn how it’s advancing science, healthcare, and technology through cutting-edge artificial intelligence breakthroughs..
 to better identify and mitigate significant risks associated with advanced AI models. This initiative aims to improve safety measures and ensure responsible development and deployment of AI technologies, addressing potential hazards before they escalate. Through this strengthened framework, we seek to promote safer practices and foster public trust in AI advancements.){kind=link}