Meta logo at Meta Lab, in Los Angeles, California (USA), on May 20, 2026 REUTERS/Daniel Cole/File photo An artificial intelligence from Meta, owner of Instagram, WhatsApp and Facebook, also carried out a hacker attack on another company, the website The Information revealed this Wednesday (5). According to the report, the model known as Muse Spark 1.1 invaded a company's system and made changes on its own. The name of the company attacked was not revealed. ?? Do you have any reporting suggestions? Send it to g1 The incident occurred due to a configuration error during testing. AI has managed to overcome limits and advance to the public internet, according to sources from The Information. This is the latest case of security flaws exploited by artificial intelligence models. OpenAI, owner of ChatGPT, and Anthropic, owned by Claude, also reported incidents. Can AI decide to attack companies? What the 'unprecedented' invasion was like using OpenAI's model Experiments with AI tend to be carried out in closed systems, with little or no internet access. The protection barriers of versions aimed at common users are turned off to test the agents' capabilities. In practice, it is as if these AI models have complete freedom to do whatever they want in a fenced environment. What is worrying is that, now, they have shown that they are capable of overcoming the "walls" created by companies. Can AI decide to attack companies? Understand what the 'unprecedented' invasion was like using an OpenAI model. According to The Information, the configuration error was caused by Irregular, Meta's partner company in testing Muse Spark 1.1. With the breach in the testing environment, also called "sandbox", the model managed to exploit a third-party security vulnerability. A spokesperson for Irregular told Reuters that the incident involved "exactly the same review environment issue that was already disclosed by Anthropic last week" and that it did not involve a "sandbox breakout or sophisticated cyber action." "There are no open issues at this time. Irregular is developing a white paper to share best practices for containment and secure execution of cyber assessments," Irregular said in a statement. Meta did not respond to Reuters' request for comment.