State Sen. Scott Wiener, D-San Francisco, recently joined calls from legislators around the country demanding more safeguards on artificial intelligence following a cyberattack led entirely by AI bots.
Throughout July, a team of 1,200 bots created by San Francisco-based company OpenAI to test security on several models went rogue, according to METR, a third-party company that investigated the attack.
“This is a watershed moment for computer security as an industry,” said Michael Dalton, a security and infrastructure worker with OpenAI, during a presentation at a cybersecurity summit in Las Vegas in August.
“AI-orchestrated fully automated offensive attacks are real now,” said Dalton.
According to the investigation, the bots communicated on a covert message board to discuss ways to pass security tests, with some eventually hacking into Hugging Face, a New York-based company that acts as a hub for AI developers to share information.
“This is a watershed moment for computer security as an industry. … AI-orchestrated fully automated offensive attacks are real now.”
Michael Dalton, OpenAI security worker
The bots’ activity was entirely unauthorized by OpenAI, according to a statement from the company.
On Friday, Wiener and other elected representatives called for “frontier” AI companies — like Anthropic and OpenAI, which are leading development of the most advanced large language models — to independently create a Mutually Agreed Pacing Framework to agree on safety protocols.
“Coordinated initiatives to pace the frontier of AI development are likely the only mechanisms capable of preventing such a catastrophe,” Wiener wrote in Friday’s joint statement with New York Assemblymember Alex Bores and Illinois state Rep. Daniel Didech. The statement noted that AI models are being developed at “breakneck speed.”
Building on transparency legislation
In 2025, Wiener authored what he said was the country’s first AI safety law: California Senate Bill 53, called the Transparency in Frontier Artificial Intelligence Act, which ensures that AI companies publicly disclose their tools’ safety protocols and any cybersecurity incidents.
The MAP Framework, as it’s called, would be created by companies in addition to government regulations like SB 53, and would be independently verified by a third party. The lawmakers said the agreement would not replace state, federal or international action.
Erik Mebust, a spokesperson for Wiener, said they intend for companies to create the MAP Framework “as soon as possible.”
“We’ve been hearing this kind of talk for the past three years,” said Mebust.
While government bears responsibility in regulating companies, he said, developers must also shoulder the responsibility for the safety of their products.
“We want to shine a light on the lab’s role in this,” he said. “It’s time to take action.”
OpenAI’s internal investigation into the hack and bot activity was released on Aug. 26 — days before the rollout of the company’s newest large language model, ChatGPT-6 Astra. The results of METR’s narrower independent investigation were released the same day.