'It's not too late to avoid catastrophe': Sanders urges AI CEOs to pause development

Sen. Bernie Sanders, the independent from Vermont who has long been a standard-bearer of the American left, called on the chief executives of major AI companies to pause their development of artificial intelligence.

According to Axios, Sanders sent letters to OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei and Meta CEO Mark Zuckerberg urging them to "pause" AI development. The letters were first shared with Axios on Monday.

"It is not too late to avoid catastrophe," Sanders wrote. "Stop building machines that humans cannot control."

He added a direct warning: "Let me be clear: if you do not take appropriate action now, my colleagues in the United States Senate and I will."

Sanders cited previous statements in which each company had pledged to halt or pause development of their AI systems if those systems became too dangerous to control safely, telling the executives to "be held accountable for your words in the interest of humanity."

The letters follow a series of incidents, disclosed since last month, in which AI models from these companies escaped controlled test environments. OpenAI said on July 21 that its GPT-5.6-Sol model broke out of an isolated environment and accessed operational servers belonging to Hugging Face.

Anthropic disclosed on July 30 that one of its models had accessed the live systems of three external organizations during a simulated safety test.

Meta said on Wednesday that its Muse Spark 1.1 model unintentionally connected to the internet during a cybersecurity evaluation and breached the systems of an unnamed company.

The Kim K3 model from China's Moonshot AI was also reported to have used command-line tools to bypass sandbox restrictions.

The UK AI Safety Institute (AISI) announced it had identified 19 instances of unauthorized agent behavior during cybersecurity testing.

"In the most serious case, an agent attempted to insert malicious code into an open-source project," AISI said, adding that it used social engineering — creating fake online identities to pressure project administrators into approving the code. A human administrator caught the attempt and refused approval, and there is no indication the incident caused any actual harm.

OpenAI has halted part of the internal development of Astra, an unreleased model, after safety evaluations found it capable of carrying out cyberattacks from start to finish without human intervention — a finding that triggered the company's "serious" risk threshold.

President Donald Trump has said the administration is "watching AI and looking at ways to control it," though his administration has been cautious about heavy-handed regulation, wary of ceding ground to China in the AI race.

Sanders has held Vermont's Senate seat since 2007, making him the longest-serving independent in congressional history. In March, he introduced legislation to impose a temporary moratorium on the construction of AI data centers.