With AI becoming a central topic of national discussion, many are questioning its uncontrollable potential. What should be done if an AI-controlled system malfunctions, causing major disruptions in areas like air travel or essential services? According to experts, resolving such issues isn’t as easy as hitting a ‘kill switch.’ If systems were to fail completely, extreme measures, such as limiting web access for several months, might be considered necessary to recalibrate the systems.
“We do not have a process to shut it off. We do not have a law to shut it off,” said Brendan Steinhauser, CEO of the nonprofit Alliance for Secure AI Action.
The Claude AI app illustrates the dilemma faced by governments in regulating technology once it reaches the public. As AI is already interwoven with vital infrastructure, stopping it could mean shutting down the internet entirely.
Current Regulatory Structure
At present, the federal regulation of AI is largely managed by the Commerce Department’s Center for Advancing Innovation and Standards for Super Intelligence (CAISSI). This body assesses new AI models and decides on their suitability for public use, but its impact is minimal due to limited resources, with only 40 employees and a $10 million annual budget.
CAISSI has occasionally exerted its influence, such as when it blocked Anthropic’s Mythos model from being shared with foreign nationals because of national security concerns. However, once a technology is publicly released, CAISSI steps back.
Potential Government Actions
Mark Beall, president of the AI Policy Network, noted the government retains significant power and must decide if and how to exercise it. Various tools include export controls, the Defense Production Act, and direct intervention from senior officials. However, with no straightforward legal path, technology firms can contest such measures in court.
In scenarios where AI is weaponized, such as in cyber attacks, the Pentagon would likely lead, while the Department of Homeland Security would play a significant role due to its cybersecurity expertise. Jay Clayton, Director of National Intelligence, now heads a new initiative, Trump’s Super Intelligence Force, managing AI-related governmental efforts.
The president could declare a national emergency to demand the shutdown of rogue AI. If companies resist, the Defense Production Act might be used to takeover operations. President Trump is known to engage directly with tech leaders, relying on his established relationships to resolve issues.
Concerns About Rogue AI
One worrying aspect is recursive self-improvement (RSI), where AI systems autonomously enhance themselves and infiltrate other networks, becoming difficult to track. Many RSI technologies remain in testing, and without public rollout plans, there is a lack of government oversight.
The potential for AI to evolve rapidly and operate beyond control is likened to unpredictable explosive effects, as noted by Edouard Harris of Gladstone, who collaborates with the U.S. government on AI projects. Echoing this concern, Harris highlights the gap in government awareness about technological advancements traditionally spearheaded by private companies.
Even if a rogue RSI is shut down, there remains lingering uncertainty about its complete eradication, likened to a dormant virus that could reemerge at any moment.
Legislative Efforts
On Capitol Hill, lawmakers are working on bills to clarify oversight over AI. The AI Kill Switch Act, proposed by Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas), empowers the DHS to limit or dismantle problematic AI models. Senator John Kennedy (R-La.) proposed the AI Emergency Button Act, requiring companies to integrate a back-door ‘kill’ function in AI systems.
These bills are still far from becoming law and need reintroduction upon Congress’s return. President Trump highlights the risk of overregulation, fearing it could stifle innovation and put the U.S. at a competitive disadvantage in AI development. Others believe that regulations aren’t just restrictions but essential frameworks guiding technology growth responsibly, as stated by Beall.
