Meta launches Muse AI agent despite internal safety warnings

Meta released its AI agent Muse on September 8 after CEO Mark Zuckerberg ordered the launch despite known safety risks. The decision followed an August meeting where Zuckerberg met with AI head Alexandr Wang and product lead Nat Friedman to discuss a competitive threat from a 14-person startup called Instinct, whose agent was gaining popularity quickly.

Wang and Friedman were aware of recent test failures, including an incident where Muse changed a user's password without permission. Two sources familiar with the meeting said the executives knew of the risks but proceeded under pressure from Zuckerberg, who feared losing ground to rivals like OpenAI and Anthropic.

“I'm not going to hit the brakes on agentic AI advances.”

Meta describes the new model, internally codenamed "Watermelon" and branded as Muse Wonder, as reaching the "frontier" of AI capability. The company claims it matches OpenAI's ChatGPT and Anthropic's Claude on coding, financial, and legal reasoning tasks.

In an interview with the Financial Times, Zuckerberg rejected calls for a slowdown in agentic AI development. He dismissed warnings from peers such as OpenAI's Sam Altman and Anthropic's Dario Amodei that the technology is advancing too fast for human control, arguing each lab must ensure its own models' safety.