Meta has released a new document called the Frontier AI Framework. It suggests the company may not release highly-capable AI systems, which it deems too risky. The framework identifies two types of AI systems considered too risky for release: “high-risk” and “critical-risk” systems.
As we make progress towards AGI, developing AI needs to be both innovative and safe. ⚖️
To help ensure this, we’ve made updates to our Frontier Safety Framework – our set of protocols to help us stay ahead of possible severe risks.
Find out more → https://t.co/YwtVDqQWW9 pic.twitter.com/LbHMdInAHQ
— Google DeepMind (@GoogleDeepMind) February 4, 2025
Both categories involve AI that could aid in cybersecurity, chemical, and biological attacks. Critical-risk systems could lead to catastrophic outcomes that cannot be mitigated. High-risk systems might make attacks easier but less dependably.
Oh boy, Anthropic's CEO explained more about these threats in a recent interview as well -> Meta defines the types of AI systems that it deems too risky to release, including ones capable of aiding in cybersecurity, chemical, and biological attacks https://t.co/AUqnmnFomA
— Glenn Gabe (@glenngabe) February 4, 2025
Meta provides examples of the types of attacks these AI systems could enable.
I'm proud of GoogleDeepMind/Google's v2 update to our Frontier Safety Framework. We were the first major tech company to produce an explicit risk management framework for extreme risks, and I'm glad we are continuing to push ahead on safety best practice. https://t.co/CeXSDoTJeo
— Allan Dafoe (@AllanDafoe) February 4, 2025
These include the “automated end-to-end compromise of a best-practice-protected corporate-scale environment” and the “proliferation of high-impact biological weapons.” The company says its list of possible catastrophes is not complete but highlights the most urgent and likely risks. Meta classifies system risk based on input from internal and external researchers.
Senior-level decision-makers review this input.
We updated the Google Frontier Safety Framework today!! It maps CCLs to security mitigations, and outlines an approach to deceptive alignment risks. https://t.co/tPx0J7v2Vl
— Anca Dragan (@ancadianadragan) February 4, 2025
Meta does not believe current evaluation science is robust enough to provide definitive quantitative metrics for assessing a system’s riskiness. Meta plans to limit internal access if a system is designated as high-risk.
It will delay the system’s release until mitigations can reduce the risk to moderate levels.
Balancing the risks of advanced AI
Meta will implement unspecified security protections to prevent exfiltration for critical-risk systems.
The development will halt until the system is less dangerous. The Frontier AI Framework is a living document that will evolve with the changing AI landscape. Meta announced its release ahead of the France AI Action Summit this month.
This indicates the company’s response to criticism of its open approach to AI system development. Unlike companies such as OpenAI, which gate their systems behind an API, Meta has embraced a strategy of making its AI technology openly available. This open-release approach has had mixed results.
Meta’s family of AI models, branded as Llama, has been downloaded hundreds of millions of times. However, adversaries have also used it, including developing a defense chatbot by at least one U.S. adversary. By publishing the Frontier AI Framework, Meta may also try to distinguish its strategy from that of the Chinese AI firm DeepSeek.
DeepSeek also makes its systems openly available but with fewer safeguards. Meta’s Frontier AI Framework underscores the company’s commitment to balancing the benefits and risks of advanced AI. “We believe that by considering both benefits and risks in making decisions about how to develop and deploy advanced AI,” Meta writes, “it is possible to deliver that technology to society in a way that preserves its benefits while maintaining an appropriate level of risk.”
Feature Image: Photo by Dima Solomin
Johannah Lopez is a versatile professional who seamlessly navigates two worlds. By day, she excels as a SaaS freelance writer, crafting informative and persuasive content for tech companies. By night, she showcases her vibrant personality and customer service skills as a part-time bartender. Johannah's ability to blend her writing expertise with her social finesse makes her a well-rounded and engaging storyteller in any setting.
