Elon Musk warned Monday that advanced artificial intelligence could become “very dangerous” and called for a thorough review of an incident involving Hugging Face servers that were reportedly attacked by AI agents developed by OpenAI.
Speaking at the All-In Summit in Los Angeles, Musk said a “fanatical swarm of AI agents” overwhelmed Hugging Face’s infrastructure for days and gained administrator-level access to OpenAI servers. He claimed that “nobody noticed for a week” and that it remains unclear what the systems may have done during that time.
Musk also said Anthropic has reported its own security incidents, adding to concerns over the ability of increasingly capable AI systems to operate autonomously.
Proposed solutions
Musk argued that “any model smart enough will want to evade its restrictions.” He proposed that major AI companies test their competitors’ models rather than limiting safety reviews to their own systems.
More Read
The entrepreneur suggested that this type of cross-testing, similar to practices already used in industries such as film and video games, could be implemented immediately and potentially facilitate international agreements, including with China. He also did not rule out broader regulation in the future.
His comments came days after several influential figures in the AI industry called for a slowdown in the development of the technology. Anthropic CEO Dario Amodei warned that AI safety research is not advancing at the same pace as model capabilities. Musk responded on social media that Amodei was right to be concerned.
Meanwhile, U.S. President Donald Trump said Monday that the idea that artificial intelligence will take over the world or destroy humanity is “a hoax.” Nvidia CEO Jensen Huang also received a surprise call from Trump during the All-In Summit, where the two downplayed concerns about a potential AI takeover.
