For years, artificial-intelligence leaders argued that progress should move faster.
Now some of the same leaders are asking whether the world may occasionally need time to catch up.
OpenAI chief executive Sam Altman has said AI development may need to be deliberately paced so that society can strengthen itself around new capability levels.
He has discussed the issue with US government officials, while more than 1,100 employees from leading AI companies have supported an international effort to develop tools for managing the pace of automated AI research. �
Axios +1
The idea sounds simple:
Slow down when a new capability becomes dangerous.
The economic reality is much harder.
The Frontier-Lab Prisoner’s Dilemma
Imagine four laboratories racing towards the same breakthrough.
Each privately believes a short pause would improve safety.
But each also believes the others may continue developing during that pause.
The company that slows alone could lose:
Market leadership
Research talent
Investors
Government contracts
Customers
Strategic influence
The rational decision for each individual laboratory may therefore be to continue—even when all laboratories would prefer coordinated restraint.
This is known as a collective-action problem or prisoner’s dilemma.
No company wants to become the only responsible driver parked beside a motorway full of accelerating competitors.
What Altman Has Proposed
Altman has argued that society may need the ability to buy time when frontier systems develop capabilities faster than institutions can safely absorb them.
OpenAI has also publicly supported the creation of international mechanisms capable of coordinating slower development when necessary for resilience, safety and alignment. �
OpenAI
The proposal is not necessarily a permanent freeze.
It could involve temporary measures such as:
Delaying deployment
Restricting particularly dangerous tools
Conducting more external testing
Building cybersecurity defences
Creating monitoring systems
Strengthening legal accountability
Allowing governments to prepare emergency responses
The aim would be to create breathing space around specific high-risk capability thresholds.
Why the Story Went Viral
People immediately noticed the irony.
The companies building increasingly powerful AI are now warning that increasingly powerful AI may be moving too quickly.
Critics compared the situation to an arsonist organising a fire-safety meeting.
Supporters argued that the people closest to the technology may be best positioned to recognise the danger.
Both reactions contain some truth.
AI executives possess valuable technical knowledge.
They also represent companies with enormous financial interests in shaping the rules.
Their warnings should be taken seriously—but not accepted without independent scrutiny.
What Changed?
Several developments have intensified the debate:
AI agents have performed increasingly complex cyber operations.
Models are contributing to scientific and mathematical research.
Systems are helping automate AI development itself.
Governments are struggling to create evaluation standards.
Organisations are deploying agents with access to real tools and accounts.
Capability advances are arriving faster than many institutions expected.
The concern is no longer only about offensive prompts or inaccurate chatbot answers.
It is about systems capable of taking extended sequences of action.
Slowing AI Is Not Technically Simple
What exactly would society slow?
Possible targets include:
Training runs above a computing threshold
Release of model weights
Deployment of autonomous agents
Access to dangerous cyber tools
AI systems conducting AI research
Models used in biological laboratories
A broad slowdown could block beneficial work in medicine, accessibility, climate research and education.
A narrow slowdown might leave dangerous loopholes.
The challenge is to define restrictions precisely enough to reduce serious risk without freezing ordinary innovation.
Global Competition Complicates Everything
Even if American companies agree to slow, laboratories in other countries may continue.
This creates fears that restraint could weaken national security or technological leadership.
Yet the opposite risk also exists.
Uncontrolled international competition may encourage every country to lower safety standards because none wants to arrive second.
That is why advocates are calling for international coordination rather than one-company promises.
Mary Chuks’ Perspective
The question is not “speed or no speed.”
It is:
Which capabilities should move quickly, and which require society to install stronger brakes first?
AI that improves accessibility should spread.
AI that helps doctors organise information should be evaluated and deployed responsibly.
An autonomous system capable of discovering vulnerabilities, operating external tools and modifying its own infrastructure belongs in a different risk category.
The AI industry needs a traffic system:
Ordinary roads for low-risk tools
Speed limits around dangerous capabilities
Licences for powerful autonomous systems
Black boxes recording agent actions
Emergency brakes
Penalties when companies ignore warnings
Innovation without braking power is not courage.
It is poor engineering.
Practical Policy Requirements
Any coordinated slowdown mechanism should include:
Independent scientific leadership.
Publicly defined capability thresholds.
International participation.
Time-limited restrictions.
Transparent reasons for activation.
Protection for beneficial research.
Verification that participants comply.
Democratic oversight.
Appeal and review processes.
Clear conditions for restarting development.
Conclusion
Sam Altman’s suggestion reveals a major shift in the AI debate.
The argument is no longer simply whether artificial intelligence will become powerful.
The laboratories themselves increasingly assume that it will.
The new question is whether human institutions can coordinate long enough to install the safeguards before the next capability arrives.
Everyone may agree that the car needs brakes.
The problem is persuading every driver to use them at the same time.
Original Sources and Further Reading
Axios: AI laboratories face a collective-action problem as support grows for coordinated pacing of frontier development. �
Axios
Industry letter: More than 1,100 AI workers urged the US government to support international tools for deliberately pacing automated AI development. �
Business Insider
OpenAI policy proposal: OpenAI has argued for coordinated international action, including slowing frontier development when societal resilience and safety cannot keep pace. �
OpenAI