U.S. Sen. Ted Cruz is under growing pressure to back stronger artificial intelligence oversight following a viral warning by a former Anthropic employee that the technology is becoming dangerously advanced.
The social media post, which has more than 150 million views, set off a wave of scrutiny not only of AI companies but also of Congress’ limited efforts so far to regulate the sector. Cruz, who has long argued for limited regulation, is positioned to help lead the response as chair of the powerful Senate Commerce, Science and Transportation Committee.
In interviews this week, Cruz described the post as “scary as hell” and said more government safeguards are needed. But he has yet to lay out specific policies or say whether he plans to hold hearings. A spokesperson declined to comment on his plans.
The post by the former Anthropic researcher, warning the technology “could kill us all by the end of the decade,” set off calls from Democrats and Republicans alike for emergency hearings and from watchdog groups that say the federal government needs to decide whether AI models are safe before they are released.
Anthropic itself released a report Thursday that detailed past efforts by “bad actors” to use its technology for cyberattacks, surveillance and potentially developing biological weapons, writing that “as models become increasingly capable, their risks will increase, unless AI developers and society’s defenders act to make them safer.”
“The ground has shifted in the last 24 hours,” said Brendan Steinhauser, a GOP consultant and CEO of the non-profit The Alliance for Secure AI. “People are worried about the loss of control, the loss of jobs, combined with the data center backlash. You’re going to see more political pressure to act.”
During an appearance on The View Wednesday, Cruz said he was working with Senate Majority Leader John Thune, R-S.D., and U.S. Sen. Amy Klobuchar, D-Minn., on legislation that would require “government supervision and approval” for AI systems that pose catastrophic risks to public safety, such as getting control of nuclear or biological weapons systems.
But he has avoided saying what level of government oversight he supports. When asked whether it would have enough support in Congress, Cruz said it was unclear and detailed concerns he and other Republicans have long expressed about AI regulation.
“One of the real challenges is [that] China is going full speed ahead, and whatever we do here, China is not going to stop,” Cruz said. “If there are going to be killer robots, I would rather they be American killer robots, rather than Chinese killer robots.”
In an interview with Elon Musk on his podcast last year, he used the same line when Musk described the chances AI would destroy humanity within the next ten years as “20% likely, maybe 10%.”
U.S. Rep. Greg Casar, a San Antonio Democrat, who is readying legislation with U.S. Sen. Bernie Sanders, a Vermont Independent, to put the development of advanced AI models on hold until more analysis can be done of the potential risks and banning so-called “superintelligent” systems altogether, says Republicans are slow-walking.
“We should be able to say there should not be autonomous killer robots for any country,” Casar said. “Trump Republicans have pushed this sociopathic policy where they think we should rush towards extremely dangerous AI because China is rushing towards extremely dangerous AI.”
Cruz has long been reluctant to put guardrails on the development of AI technology, and instead focused his attention on the spread of so-called deepfake pornography, images in which people are falsely made to appear as if they are having sex.
In a Senate hearing last year, Cruz called for a “light-touch regulatory style for AI,” criticizing the Biden administration’s set of rules as likely to give China the upper hand in the global AI race. So far this year, Cruz’s Senate Commerce Committee has held hearings on AI topics including how the technology can be used to tighten security of communications networks and “improve Americans’ quality of life, create jobs, and drive economic growth.”
His approach is in line with other Republicans, who since President Donald Trump took office last year, have been focused on a hands-off approach to AI, doing away with many protections established by the Biden administration, said Cobun Zweifel-Keegan, a managing director at IAPP, a Washington non-profit studying data and artificial intelligence. Among the policies done away with by Trump were an executive order requiring the testing of AI models to ensure they don’t create biological or nuclear weapons and a Commerce Department rule limiting the export of advanced AI technology.
“The Republican approach has been let’s focus on innovation and worry about guard rails later if we need to,” he said. “Cruz has embraced and led on those issues.”
But the discussions around AI are quickly evolving.
Even before the Anthropic employee’s warning, OpenAI disclosed last month that unbeknownst to developers, one of it’s AI models had hacked into another company’s system and managed to keep its efforts hidden for weeks. In a letter to Trump, nonprofits monitoring AI technology warned that incident and other recent breaches proved new AI technologies can “behave in ways their developers do not intend and that existing safeguards may be insufficient to prevent.”
States including New York, Illinois and California have already passed legislation requiring AI companies to notify them if their models show signs of activity that could endanger human lives or cause serious economic harm.
Similar legislation is under discussion in the U.S. House, and U.S. Rep. Nathaniel Moran, an East Texas Republican, introduced a bill last month with U.S. Rep. Ted Lieu, a California Democrat, requiring AI companies to maintain a “kill switch” should their systems show signs of threatening human lives.
But figuring out a system to safeguard against cataclysmic actions by AI systems is likely to take time. Even among those who have been tracking the technology’s progress closely, widespread debate remains over what protections are needed, said Zweifel-Keegan.
“We don’t really know what we’re dealing with,” he said. “It’s hard to legislate things that haven’t occurred before.”