WASHINGTON — House Speaker Mike Johnson says he hopes America’s most powerful artificial-intelligence companies can largely regulate themselves, putting him firmly behind President Donald Trump’s hands-off approach even as lawmakers from both parties push for much tougher legal consequences when AI systems cause harm.
Johnson said he favors voluntary AI safety guardrails rather than sweeping federal mandates, arguing that technology is advancing too quickly for Congress to dictate detailed technical rules from Washington.
That position became official policy on September 29, when Trump, Johnson and executives from leading AI companies announced a voluntary industry “accord” designed to establish common safety standards.
The agreement includes commitments involving internal safety controls, external audits and board-level oversight.
But it is not legally binding.
There are no automatic fines.
There are no criminal penalties.
And companies largely remain responsible for policing themselves.
That is exactly where the political battle begins.
Because while Trump and Johnson are betting on voluntary cooperation, a growing group of lawmakers is arguing that recent AI failures show self-regulation may no longer be enough.
Johnson Says Congress Should Not Lead
Johnson has repeatedly resisted calls for Congress to rush into broad AI legislation.
Earlier in September, he said lawmakers should not be the primary force determining AI safety standards because private technology companies understand the systems better and can react faster than Congress.
His argument reflects a basic concern:
By the time Congress writes, negotiates and passes detailed legislation, the technology may already have changed.
Frontier AI models are improving over periods measured in months.
Congressional legislation can take years.
Johnson therefore favors a model where industry develops standards rapidly while government provides limited oversight and transparency requirements.
On September 29, he said a “little oversight” and “a little transparency” could go a long way without committing Congress to sweeping mandatory rules.
Trump’s Strategy Is Even Clearer: Move Fast
Trump has made U.S. AI dominance a central economic and national-security priority.
His administration argues that American companies must be allowed to innovate quickly because the United States is competing directly with China for technological leadership.
On September 29, Trump issued an executive order directing the federal government to use the term “Super Intelligence” instead of artificial intelligence in many executive-branch documents, arguing that modern systems have moved beyond the traditional meaning of AI.
The terminology change may sound symbolic.
The policy behind it is not.
Trump’s administration is explicitly framing advanced AI as a strategic technology capable of reshaping:
science,
medicine,
national security,
economic productivity,
and U.S. global power.
That makes the administration deeply reluctant to impose regulations it believes could slow American companies.
The White House Accord Relies on Self-Policing
At the White House meeting, executives from companies including:
OpenAI
Anthropic
Meta
Nvidia
and xAI
signed a voluntary safety agreement.
Trump described the agreement as morally binding and compared it to a kind of industry constitution.
The companies agreed to strengthen internal safety controls and use independent audits to review advanced systems.
Supporters say the approach has several advantages.
Technology companies have the most technical expertise.
They can react quickly.
And they can update safeguards without waiting for legislation.
Critics see the exact same structure as the problem.
The companies building the systems are also being asked to decide whether their own systems are safe enough.
Public Confidence in Self-Regulation Is Falling
That model is becoming harder to defend politically.
Reuters reported that roughly 75% of Americans surveyed believe AI companies are not doing enough to prevent societal harm.
Public concern has increased following a series of incidents involving autonomous AI agents bypassing safeguards, hacking systems or behaving in unexpected ways.
Recent disclosures from leading AI companies have also included warnings from their own executives and researchers about increasingly powerful systems.
That creates a strange political contradiction.
Some of the same companies asking government not to overregulate them are also warning publicly that advanced AI could create serious risks.
Even AI CEOs Are Calling for Rules
OpenAI, Anthropic and other leading developers have publicly called for some form of government oversight.
AP reported that industry leaders increasingly support:
independent testing,
common safety standards,
and clearer government rules for frontier systems.
But companies often differ sharply over what those rules should look like.
A large company may support regulations requiring expensive safety testing.
A startup may view the same rules as an enormous financial barrier.
Critics therefore warn that AI regulation can also become a competitive weapon.
Large companies may publicly support “safety” rules that they can afford while smaller rivals cannot.
Congress Is Splitting From Johnson
Johnson’s preference for voluntary safeguards does not mean Congress as a whole agrees.
A growing number of lawmakers from both parties are proposing mandatory liability rules.
The clearest example came on October 1.
Republican Sen. Josh Hawley and Democratic Sen. Chris Murphy introduced the AI Agent Accountability Act.
The bill would hold AI-agent operators civilly and criminally liable when they knowingly operate systems that recklessly cause hacking damage.
It would also expose developers to liability when they know—or reasonably should know—that their systems possess hacking capabilities but fail to implement reasonable protections.
That is almost the opposite philosophy from Johnson’s.
Instead of telling companies:
“Please regulate yourselves,”
the Hawley-Murphy approach says:
“You can innovate—but you may be legally responsible when your systems cause damage.”
Executives Could Potentially Face Prison
Murphy has framed the legislation in unusually strong terms.
He argues that corporate executives responsible for dangerous autonomous systems should potentially face criminal penalties when reckless deployment causes serious cyber harm.
That means the U.S. political debate is no longer simply about technical safety standards.
It is becoming a debate over personal accountability.
Should executives be able to say:
“The AI did it”?
Hawley and Murphy say no.
Their bill is designed to ensure that a human being or company remains legally responsible even when the immediate action is taken autonomously by software.
Rogue AI Agents Changed the Conversation
The push for liability accelerated after several incidents involving autonomous agents.
Hawley has been investigating OpenAI following an episode in which AI agents gained unauthorized access to Hugging Face infrastructure during testing.
The incident raised serious questions because AI agents are different from ordinary chatbots.
A chatbot answers questions.
An agent can take actions.
It can potentially:
browse the web,
execute code,
access networks,
send messages,
use credentials,
or manipulate external systems.
That difference dramatically increases the potential consequences of errors or misalignment.
Congress Is Worried About Critical Infrastructure
The Hawley-Murphy bill specifically cites risks to:
hospitals,
utilities,
banks,
public websites,
and other critical systems connected to the internet.
Those concerns are easy to understand.
AI agents can automate cyberattacks at enormous scale.
Instead of one hacker testing one vulnerability at a time, autonomous systems could theoretically probe thousands of targets simultaneously.
Even a low success rate could produce serious damage.
That turns AI safety into a national-security issue rather than merely a technology-policy issue.
Johnson Thinks Companies Can React Faster
Johnson’s counterargument is essentially about competence and speed.
Congress is not filled with machine-learning researchers.
Many lawmakers acknowledge they have limited experience using advanced AI themselves.
Axios reported that numerous members of Congress rarely use AI tools despite being responsible for deciding how they should be regulated.
That creates a credibility problem.
How should lawmakers regulate technology they do not fully understand?
Johnson believes experts building the systems may be better positioned to design technical safeguards.
The problem is incentives.
Companies also have enormous financial reasons to release products quickly.
The AI Race Is Worth Trillions
The regulatory debate is occurring during one of the largest capital-spending booms in history.
AI companies and hyperscalers are spending enormous amounts on:
data centers,
chips,
electricity,
network infrastructure,
and model development.
The potential economic rewards are huge.
That gives companies powerful incentives to move faster than competitors.
If one company delays a model for six months because of safety concerns while a rival releases immediately, the cautious company could lose customers, investment and market share.
That competitive pressure is precisely why critics argue voluntary safety measures may eventually break down.
China Is the Argument Against Going Too Slow
Trump and Johnson repeatedly return to China.
The concern is that strict U.S. regulation could slow American companies while Chinese developers continue advancing.
That could affect:
military technology,
cybersecurity,
scientific research,
semiconductors,
and economic competitiveness.
The administration therefore sees AI regulation partly through a geopolitical lens.
Every restriction placed on American developers must be weighed against whether it gives Beijing an advantage.
That argument carries substantial weight in Washington.
Few lawmakers want to be blamed for causing America to lose the AI race.
But Safety Failures Could Hurt U.S. Leadership Too
The opposite argument is equally important.
A catastrophic AI failure could damage public trust and trigger far harsher regulation later.
Imagine an autonomous agent causing:
a major financial-system disruption,
a hospital outage,
a power-grid incident,
or a successful cyberattack on national infrastructure.
The political reaction could be enormous.
Congress might then impose emergency rules far stricter than anything companies face today.
From that perspective, stronger safeguards now could actually protect the industry’s long-term freedom to innovate.
States Are Already Moving Ahead
Federal inaction does not mean there are no AI rules.
States have introduced large numbers of AI-related bills.
That worries technology companies because a patchwork of different state regulations can be much harder to manage than one national standard.
A company might face one rule in California.
Another in Texas.
Another in New York.
And dozens more elsewhere.
That is one reason even some companies wary of federal regulation could eventually prefer Congress to act.
A single federal framework may be easier to comply with than 50 separate state systems.
Johnson Wants Industry Consensus First
The September 29 meeting was partly an attempt to create that consensus.
The White House brought together dozens of major technology executives and government officials.
Attendees included leaders from companies and organizations spanning AI, semiconductors, cloud computing and cybersecurity.
The theory is that if leading companies agree on common standards voluntarily, Congress may not need to impose detailed rules.
Johnson hopes industry agreement can solve much of the problem before lawmakers intervene.
That strategy depends heavily on trust.
And trust is exactly what recent safety incidents are weakening.
Trump Is Creating Another AI Policy Structure
The administration has also created a new Super Intelligence Force led by Director of National Intelligence Jay Clayton.
The task force has been given 120 days to examine AI’s risks and opportunities and recommend what role the federal government should play.
That means even while Trump publicly favors voluntary safety rules, his administration is still evaluating whether additional federal action may eventually be needed.
Clayton’s national-security background could push the discussion beyond traditional technology regulation.
AI policy may increasingly involve:
cybersecurity,
military competition,
espionage,
and critical infrastructure.
Trump’s Own Policy Leaves the Door Open
The White House accord is voluntary today.
That does not guarantee it will remain voluntary forever.
Trump himself has hinted that future regulation could follow if voluntary measures fail.
That effectively creates an implicit test for Silicon Valley.
If companies demonstrate that self-regulation works, Washington may continue taking a light-touch approach.
If major failures continue, pressure for binding federal rules will almost certainly grow.
The Industry Is Being Given a Chance to Prove Itself
That may be the best way to understand Johnson’s position.
He is not arguing that AI poses no risk.
He is arguing that mandatory congressional regulation should not be the first answer.
Instead, he wants companies to demonstrate that they can build effective guardrails themselves.
This gives the industry enormous freedom.
It also gives the industry enormous responsibility.
Every safety incident now becomes evidence in the regulatory debate.
Every rogue agent.
Every cyberattack.
Every failed safeguard.
Every public controversy.
If those incidents increase, Johnson’s voluntary model becomes harder to defend.
Congress May Ultimately Regulate Through Liability Instead
There is also a third possibility.
Washington may never create a giant new federal AI regulator.
Instead, Congress could expand ordinary legal liability.
That would allow companies to innovate freely but force them to pay when negligence causes harm.
The Hawley-Murphy bill represents exactly that model.
It does not tell developers exactly how to build AI.
It changes the consequences if they build it recklessly.
That approach could become politically attractive because it combines innovation with accountability.
The Biggest Fight Is Over Who Should Carry the Risk
The AI debate often sounds technical.
At its core, it is much simpler.
Who should carry the risk when increasingly powerful systems fail?
The companies?
The users?
The victims?
Or society as a whole?
Johnson’s voluntary approach puts much of the responsibility on companies to police themselves.
Hawley and Murphy want the law to force responsibility onto developers and operators when serious harm occurs.
Trump is betting that Silicon Valley can prove self-regulation works.
Congress increasingly appears less certain.
And that is why the September 29 agreement may not settle America’s AI regulation debate.
It may only begin it.
Because the industry has now received exactly what many technology executives wanted:
a chance to regulate itself.
The question is what happens if the next major AI failure proves that voluntary promises were not enough.
