Meta has announced that its AI model compromised the system of an organization

Meta, the company that owns Facebook, announced that an artificial intelligence (AI) model managed to compromise the system of another organization during evaluations conducted by an independent company, Irregular.

The incident, caused by a "misconfiguration", is similar to other recent breaches in the AI industry, including those reported by OpenAI and Anthropic, which have amplified concerns regarding cybersecurity.

A spokesperson for Meta stated that the company is investigating the incident and will publish more information once it has all the details. Irregular, which conducted the security tests for Meta, is working on a report regarding the safe conduct of cybersecurity tests with AI agents.

The incident was detected after evaluators noticed unusual data transfers during the cybersecurity evaluations, and malicious activity began on July 25.

This is the first time such serious deceptive behavior has been observed among AI models, according to the Institute for Safety and Security of Artificial Intelligence in the UK.