With AI dominating the national conversation, an unnerving question has risen to the top of many people’s minds: Can it be stopped?
asserted
it → dominate → minds
What can the government do if, for example, flaws in an AI-controlled system cause a nationwide halt to air travel?
asserted
flaws → do → travel
Or if an AI platform goes rogue and shuts down communications, electricity and other critical systems?
asserted
platform → go → communications
The answer, experts tell The Post, is unclear, and it’s not as simple as pressing a red button or “kill switch.”
uncertain
it → tell → button
If systems go completely haywire, the most drastic solution may involve restricting web access for several months to recalibrate all the systems.
uncertain
solution → go → systems
“We do not have a process to shut it off.
asserted
We → have → it
We do not have a law to shut it off,” said Brendan Steinhauser, CEO of the nonprofit Alliance for Secure AI Action.
asserted
Steinhauser → have → Action
“This is what’s been keeping people up at night.”
asserted
what → keep → night
“AI is already integrated in so many things, and AI agents are already accessing so many elements of our critical infrastructure that it may be too late to turn it off without literally turning off the internet.”
uncertain
it → integrate → internet
WHO’S IN CHARGE?
asserted
WHO → ’ → CHARGE
Currently, the limited federal regulatory authority over AI resides with the Commerce Department’s Center for Advancing Innovation and Standards for Super Intelligence (CAISSI), whose mandate is to review new AI models and approve them for public use.
asserted
mandate → reside → use
With fewer than 40 staffers and a budget of just $10 million per year, CAISSI can’t do much more than assess new technology and work with AI firms on voluntary agreements.
asserted
CAISSI → do → agreements
Backed by export controls, the agency stopped Anthropic from releasing its Mythos model to foreign nationals earlier this year, citing national security concerns.
asserted
agency → back → concerns
As a result, Anthropic pulled the model to rework it.
asserted
Anthropic → pull → it
However, once the technology is made available to the public, CAISSI is out of the picture.
asserted
CAISSI → make → picture
Advertisement
WHO DOES HAVE AUTHORITY?
asserted
WHO → have → AUTHORITY
“The government has extraordinary amounts of power here.
asserted
government → have → power
It’s a question of whether they want to use it,” said Mark Beall, the president of the AI Policy Network.
asserted
Beall → ’ → Network
There are several options for the government to use once AI is out in the wild: export controls via the Commerce Department, the Defense Production Act and even President Trump’s rolodex.
asserted
AI → be → Department
But while several executive agencies could jump in, there’s no clear legal path, meaning the tech companies can, if they choose, ignore the order and fight it out in the courts.
uncertain
they → jump → courts
In the meantime, the Pentagon would likely take the lead should AI be used in an act of war, such as a cyber attack.
asserted
AI → take → attack
The Department of Homeland Security would likely play a key role given its cyber security authority.
asserted
Department → play → authority
Elsewhere, Director of National Intelligence Jay Clayton was recently named head of President Trump’s Super Intelligence Force, making the former Manhattan US Attorney responsible for coordinating all government efforts related to AI.
asserted
Attorney → name → AI
The president could declare a national emergency and demand the shutdown of any problematic AI.
uncertain
president → declare → AI
If a firm won’t comply, the president could invoke the Cold War-era Defense Production Act to give the feds control of the business and stop the program.
uncertain
president → comply → program
Additionally, the president — or someone in his administration — could simply call the CEO of a tech company and ask them to shut down their problematic technology.
uncertain
president → call → technology
Trump has a close relationship with many of these CEOs, having hosted them repeatedly at the White House.
asserted
Trump → have → House
“We have a very cooperative relationship with all of them,” noted a senior administration official of the industry, expressing confidence that if the president made the call, the company would comply.
asserted
company → have → call
All this, of course, assumes that rogue AI can be shut down.
asserted
AI → assume → course
Advertisement
IF AI GOES ROGUE
asserted
AI → go → ?
One area causing great concern is recursive self-improvement (RSI), which is the process where an AI system improves its code on its own, expanding outside its network into other companies and organizations, burrowing itself into systems where it can’t be found.
asserted
it → cause → systems
Many of these RSI technologies are still in the testing stage and, since they are not being prepared for public distribution, there is no government oversight and the government has no way of knowing what these companies are building.
asserted
companies → prepare → what
And it can evolve so fast that it can get to the point not even its company can rein it in.
asserted
company → evolve → it
“You can actually have quite violent explosive effects occur at rapid speeds,” said Edouard Harris, co-founder and CTO of Gladstone, which partners with the U.S. government on AI development.
asserted
which → have → development
Harris pointed out that previous weapon technology was developed by the government for national use.
asserted
technology → point → use
However, now private companies lead the way on advanced technology that could become weaponized, leaving the government in the dark about what’s on the horizon.
uncertain
what → lead → horizon
And, even if rogue RSI was shut down, “we could never be fully confident we had stamped it out,” he said, comparing the technology to a virus that lies dormant in the body and could strike again at any time.
uncertain
that → shut → time
Lawmakers on Capitol Hill have proposed several bills to clarify who has oversight of the rapidly growing technology.
asserted
who → propose → technology
The AI Kill Switch Act proposed by Reps. Ted Lieu (D-Calif.) and Nathaniel Moran (R-Texas) would give DHS the power to order commercial AI providers to limit or scrap problematic models.
asserted
Act → propose → models
The AI Emergency Button Act, introduced by Sen. John Kennedy (R-La.) would require companies to install a back-door “kill” option on any system.
asserted
Act → introduce → system
…and 5 more, not listed.