‘We Are So Far Behind’: Senators Probe Experts on Risks of ‘Rogue AI’

At a hearing on “Rogue AI” agents on Wednesday, senators probed experts about how the United States can protect Americans from the risks of rapidly developing artificial intelligence technologies. Though several options around a development slowdown, a change to the legal landscape, and the creation of stronger security measures, it became increasingly clear that Congress may simply be too slow to tackle the challenges that will come in the next few years. Senator Josh Hawley, the populist conservative from Missouri, held the hearing, “Rogue AI: Securing the Homeland Against AI Agent Attacks,” to hear from safety experts about what can be done. Mr. Hawley invited OpenAI chief executive Sam Altman to appear as a witness, though Mr. Altman declined to do so. Mr. Hawley held the hearing as chairman of the homeland security subcommittee on disaster management. At the outset, the stakes of the hearing were quite clear. Paul Ohm, a Georgetown University law professor and expert in the intersection of AI and the law, offered his assessment of recent security issues at OpenAI where agents went rogue and hacked other websites without permission. Mr. Ohm said that if the reports on the agent attacks had said “OpenAI employee” instead of “AI agent,” then the reports would have read like a criminal indictment.“There is an important role to be played by criminal law,” Mr. Ohm said in his opening statement, urging Congress to work piece-by-piece in order to try to keep pace with the development of AI in order to protect Americans.“They are turning to you — their elected representatives — for answers,” Mr. Ohm said. “In conversations that can too often seem insular and circular and stalled, new points of view may be just what we need.”Daniel Kokotajlo, a former OpenAI employee who refused to sign a nondisparagement clause in order to publicly raise concerns about the company, also testified. In his opening statement, Mr. Kokotajlo said the world is barreling toward the development of an entirely independent system of AI agents that may end up wiping out humanity. He and his team estimate that the development of such technology could come “any year now,” though he estimates that there is a 50 percent chance that such development occurs by the end of 2028. He criticized three of the world’s most famous AI executives — Dario Amodei, Elon Musk, and Mr. Altman — all of whom have called for a global slowdown in AI development. “They have not paced the frontier yet,” Mr. Kokotajlo told senators, referring to the industry phrase for slowing development of agents. “I think that the U.S. government should intervene.” He warned, though, that all of the incentive has to do with making money and preparing to go public. “If a company tries to go slower and figure out how AI agents work — improve their training environments, for example … Well, that takes time, and it means the launch of the next model may be delayed and then they might fall behind other companies,” Mr. Kokotajlo said. A third expert, Marius Hobbhahn, told Senator Andy Kim — the top Democrat on the disaster management subcommittee — that it is “unknown” if humans can develop a “sandbox” that can actually contain AI agents. “It’s very, very hard to develop fully water-tight sandboxes,” Mr. Hobbhahn warned. “Even if the humans think that the sandbox is fully [secure], it could still be possible that a smarter model in the future could be able to find a way out of it.”The chairman of the Senate’s Homeland Security and Government Affairs Committee, Senator Gary Peters, attended the hearing to question witnesses. Mr. Peters, who is retiring at the end of the year, was concerned not just with the technology itself, but with Congress’s inability to move on major issues in a timely fashion. He says the Senate’s slow nature is especially dangerous as AI develops exponentially.“What is certainly a huge challenge for us … is that throughout the Industrial Revolution, technology moved much quicker than policies,” Mr. Peters said. “Now, we’re in this new world [where] technology is moving faster than older technology at an exponential rate. So, it’s not even policy. We are so far behind.” “Every time we build a guardrail, we’re not even sure if we can stop” the new models, Mr. Peters said. Senator Ruben Gallego, a possible 2028 White House contender, questioned witnesses about the threat of China developing more powerful agents that could somehow be directed to attack the United States if such a conflict were to ever arise. Mr. Gallego asked if China could somehow develop a sandbox that protects the Chinese people while endangering Americans. One witness responded that it is likely that no such safety mechanism exists. “That should scare the shit outta people,” Mr. Gallego responded.

‘We Are So Far Behind’: Senators Probe Experts on Risks of ‘Rogue AI’

‘We Are So Far Behind’: Senators Probe Experts on Risks of ‘Rogue AI’

More from the New York Sun

A Disillusioned Democrat Burns Her Bridges, Seeking To Light the Way For the Left

Justice Department Seeks To Disqualify Biden-Appointed Judge Over ‘Animosity for the Trump Administration’

Iran Dials Up Pressure on Hormuz With Record Number of Weekly Strikes on Shipping

‘Beyond Irresponsible’: Republicans on Capitol Hill Denounce Trump’s Decision To Waive Russia Sanctions

Out of Adversity, a Japanese American Design Vision

The Tesla Roadster Can Really Fly, But It’s Not a Real Car Yet

Blakeman Hopes To Cut a Path to Albany Through Minority and Jewish Voters, and Hochul’s Unpopular Policies

With 20,000 Names Struck From Michigan’s Sex Offender List, Victim Rights Advocates Wonder What Good It Is

More from the New York Sun

A Disillusioned Democrat Burns Her Bridges, Seeking To Light the Way For the Left

Justice Department Seeks To Disqualify Biden-Appointed Judge Over ‘Animosity for the Trump Administration’

Iran Dials Up Pressure on Hormuz With Record Number of Weekly Strikes on Shipping

‘Beyond Irresponsible’: Republicans on Capitol Hill Denounce Trump’s Decision To Waive Russia Sanctions

Out of Adversity, a Japanese American Design Vision

The Tesla Roadster Can Really Fly, But It’s Not a Real Car Yet

Blakeman Hopes To Cut a Path to Albany Through Minority and Jewish Voters, and Hochul’s Unpopular Policies

With 20,000 Names Struck From Michigan’s Sex Offender List, Victim Rights Advocates Wonder What Good It Is