Meta recently released a new risk policy framework aimed at assessing and mitigating the risks posed by cutting-edge AI models, and, if necessary, halting development or restricting the release of these systems. This framework, called the "Cutting-Edge AI Framework," outlines how Meta will categorize AI models into high-risk and critical-risk categories and take appropriate measures to reduce risks to "tolerable levels."
In this framework, critical risks are defined as those that can uniquely contribute to the execution of specific threat scenarios. High risks, on the other hand, indicate that the model may significantly increase the likelihood of realizing threat scenarios but does not directly contribute to their execution. Threat scenarios include the proliferation of biological weapons that could match known biological agents, as well as widespread economic harm to individuals or companies caused by large-scale long-term fraud and scams.
For models that reach the critical risk threshold, Meta will halt development and restrict access to the model to a select few experts, while implementing safeguards to prevent hacking or data breaches, provided it is technically and commercially feasible. For high-risk models, Meta will limit access and take measures to mitigate risks, aiming to reduce the risk to a moderate level, ensuring that the model does not significantly enhance the execution capability of threat scenarios.
Meta stated that its risk assessment process will involve multidisciplinary experts and internal leadership to ensure that all perspectives are adequately considered. This new framework applies only to the company's state-of-the-art models and systems, which match or exceed current technological levels.
Meta hopes that by sharing its development approach for advanced AI systems, it can enhance transparency and promote external discussions and research on AI assessment and risk quantification science. The company also emphasizes that the decision-making process for AI assessment will evolve and improve with technological advancements, including ensuring that the results of its testing environment accurately reflect the model's performance in real-world operations.
Key Points:
🌟 Meta introduces a new risk policy framework to assess and mitigate risks of cutting-edge AI models.
🔒 Critical risk models will halt development and limit expert access; high-risk models will implement access restrictions and mitigation measures.
🧑🏫 The risk assessment process will involve multidisciplinary experts to enhance transparency and scientific rigor.
