Mark Zuckerberg, the CEO of Meta, has made a promise to one day share a special kind of artificial intelligence (AI) called artificial general intelligence (AGI) with everyone. AGI is the kind of AI that can do any task that a human can do. However, in a new set of guidelines, which Meta refers to as the Frontier AI Framework, the company has indicated that there are specific situations where it might hesitate to release certain powerful AI systems it has created.
In this document, Meta highlights two categories of AI systems that it considers to be too dangerous to be made public: “high risk” and “critical risk” systems.
Both types of systems have the potential to assist in harmful activities, like cybersecurity attacks or creating dangerous chemical and biological weapons. The key distinction is that “critical-risk” systems could lead to outcomes that are disastrous and cannot be controlled or stopped in the context they would be used in. On the other hand, “high-risk” systems could make it easier to carry out an attack, but they wouldn’t be as reliable or effective in doing so as critical-risk systems.
For example, some of the dangerous scenarios that Meta has in mind include situations like completely taking control over a well-protected company’s computer environment without permission or the spread of powerful biological weapons that could harm many people. While these examples are just a few, Meta acknowledges that there are many other serious risks that could arise if a powerful AI system was released into the world.
Interestingly, rather than relying on a single test to decide how risky a system might be, Meta determines risk based on the insights from both its own researchers and outside experts. These inputs are then reviewed by higher-level decision-makers within the company. Meta explains that they do not think the current scientific methods for evaluating AI are strong enough to provide clear and accurate measurements of risk levels.
Read Also: DeepSeek: Countries And Agencies Banning The AI Company’s Technology
If Meta decides that a particular system is high-risk, the company plans to limit who can access that system internally, and it won’t share it publicly until it has taken steps to lower the risk to a more manageable level. If a system is classified as critical-risk, they will put certain security measures in place to ensure that it cannot be leaked or misused and will stop developing it until they can figure out how to make it safer.
The Frontier AI Framework is something that Meta says will adapt as the field of AI evolves. Meta has made this commitment public ahead of an upcoming AI summit in France, as a response to some criticisms they’ve faced about their approach to AI development. The company has generally been more open about sharing its AI technology, unlike other companies, such as OpenAI, which keep their systems more locked down and available only through controlled access.
While this open approach has allowed Meta’s group of AI models, called Llama, to be downloaded hundreds of millions of times, it has also led to some complications. For instance, there was a reported case where someone in the U.S. used Llama to create a chatbot for defense purposes.
By sharing this Frontier AI Framework, Meta might also be trying to highlight how their strategy differs from that of a Chinese AI company called DeepSeek. Like Meta, DeepSeek makes its systems available to the public, but their AIs lack adequate safeguards, making it easy for users to coax them into producing harmful or problematic content.
In the document, Meta asserts, “We believe that by considering both benefits and risks in making decisions about how to develop and deploy advanced AI, it is possible to deliver that technology to society in a way that preserves the benefits while also managing the level of risk.” Essentially, they want to find a balance between making powerful AI available and ensuring that it doesn’t lead to catastrophic consequences.
