Goal establishes a framework to identify and mitigate “possible catastrophic results” derived from AI systems

Meta explained how he works to build artificial intelligence tools and systems (AI) advanced, in a frame “derived from the use and development of this technology.

In a document, called Frontier AI Framework, it has described how it works to build advanced and how it establishes catastrophic thresholds; An approach that expects “evolving and maturing over time”, as the uses of this technology progress.

With this, the company has underlined its commitment to “advance in the last in AI, in models and systems to implement them in a responsible manner”; And he has recognized that, although “It is not possible to completely eliminate the risk“Intrinsic that has AI, it does seek to be” positive for society. ”

In this way, he has indicated that his work framework ‘high risk’ and of ‘critical risk’being the latter those related to the areas of threats to cybersecurity, as well as those of chemical and biological weapons of high impact “.

Frontier AI Framework, therefore, has been established to identify possible catastrophic results and evaluate whether these results “are possible thanks to technological advances.” “If so, We will identify ways of alleviating these risks“He pointed out.

Meta has also advanced that performs threat modeling exercises to anticipate how different actors could try to make improper use of AI to produce the commented catastrophic resultsworking with external experts as necessary.

He has also advanced that he defines risk thresholds depending on the degree to which their AI models facilitate threat scenarios. To do this, it has established processes to maintain the risks within acceptable levels, including the application of mitigation measures.

In case of identifying that an AI system is High riskIt will limit its access internally until it has completed the relevant mitigation deployment to “reduce risk to moderate levels.” Before a Critical riskand if it is considered that an AI has reached the established threshold and cannot be mitigated, goal has stressed that it will stop its development and implement other protections, although it has not specified which ones.

Technology has finally pointed out that although the objective of this framework is to reduce serious risks, the main reason why these technologies are developing “is because they have the potential to greatly benefit society.”

It should be noted that the announcement of this framework goes in the line of the Artificial Intelligence Law of the European Union, the pioneering legislation to regulate this technology worldwide, which entered into force on August 1 and aims to guarantee security and the fundamental rights of EU citizens.

This regulation seeks to regulate its low use A uniform legal frameworkfacilitating the marketing and circulation of products and systems based on AI, not forgetting cybersecurity and technological development under an ethical approach by the companies that develop it.

It also has an approach based on the risk levels of presenting the AI, which are collected in four categories: The systems that suppose an unacceptable risk -which are directly prohibited -, high -risk systems, limited risk systems and minimum risk systems. In this category are most AI systems.

Although the mandatory application of these regulations will begin in August 2026, it has been this Sunday when the prohibitions of artificial intelligence systems have been applied that are considered to have an “unacceptable risk”, at which time the goal has taken advantage of to explain which It is your current approach to the responsible work of advanced systems of AI.