Former President Barack Obama and former Vice President Kamala Harris have joined a rapidly widening debate over whether the development of the most advanced artificial intelligence systems should be slowed and more tightly supervised.
Their interventions came after Anthropic chief executive Dario Amodei published a proposal calling for “pacing the frontier” —not a permanent halt to AI research, but a slower, more verifiable process in which safety work keeps pace with capability gains.
OpenAI chief executive Sam Altman, Tesla and SpaceX chief Elon Musk, and Google DeepMind chief Demis Hassabis have expressed support for parts of that approach.
The political stakes are high. Supporters argue that increasingly capable AI systems may create risks in cybersecurity, biological research, labor markets and democratic institutions before governments and independent evaluators can respond.
Critics warn that a broad slowdown could strengthen the largest incumbent laboratories, weaken open-source competitors, and allow China to overtake the United States in a strategically important technology.
What Obama and Harris said
Obama’s September 14 post on X was authentic and closely matches the text supplied in the question.
He said he was encouraged that frontier-lab leaders had agreed on the need to slow development and argued that AI’s future should be a subject of public debate rather than a decision left principally to private companies.
He rejected both “accelerationist” optimism and “doomer” fatalism, while warning that AI could produce medical, energy and educational gains—or economic disruption, inequality and
catastrophe—depending on choices made now.
The post followed remarks Obama made at a Democratic Congressional Campaign Committee fundraiser, where he urged Democrats to develop a public framework covering safety, children’s wellbeing and job displacement.
He said AI was moving quickly in private hands and could be dangerous if society failed to “get on top of it,” while also emphasizing its potential to accelerate drug development and clean-energy research.
Harris’s September 14 X post was also authentic. She called the pace of frontier -AI development “alarming” and said a slowdown was necessary to ensure that AI served the public interest.
She emphasized that pacing did not mean abandoning beneficial applications such as medicine.
Instead, she called for congressional action, a new federal entity with responsibility for oversight and independent testing, and an international treaty with countries including China to restrict dangerous uses, limit the speed of development and establish common testing standards.
Neither politician called for ending AI research. Both framed the issue as one of democratic control, independent scrutiny and risk management.
What “pacing” would mean in practice
Amodei’s proposal is more specific than a general appeal for caution. He proposes that frontier laboratories give embedded third-party evaluators ongoing, employee-like access to models, training pipelines and safety processes.
The evaluators would verify safety
commitments, report incidents and assess systems before and during deployment.
Anthropic says it is committing to this step unilaterally. His second proposal is coordination among frontier companies in democratic countries to establish common safety standards and limits on unchecked progress.
His third is global coordination, including difficult discussions with authoritarian governments. Amodei says pacing should not halt model training or technical progress.
It should instead require companies to spend sufficient time aligning and safeguarding systems, with independent evaluators checking whether they have done so. That distinction matters because a slowdown could target several different things.
It could delay the release of systems that cross defined capability thresholds, restrict the scale of training runs, require safety certifications, or slow deployment of autonomous agents.
It could also mean pausing specific research programs after an incident rather than imposing a blanket moratorium.
The industry reaction is unusually broad—but not unanimous Altman wrote that AI progress should be slower than it otherwise could be and endorsed
independent evaluators.
Musk said Amodei was right. Hassabis described the direction as appropriate to the moment.
Hugging Face chief executive Clément Delangue supported more transparent safety work and argued that the problem should not be solved solely behind the closed doors of a small number of frontier laboratories.
The apparent consensus has limits. Some critics suspect that major labs could use safety regulation to raise rivals’ costs and consolidate control over advanced models.
Investor Chamath Palihapitiya argued that Amodei’s plan could concentrate technological and economic power in Anthropic and other closed-model companies.
David Sacks, a technology investor and former White House AI official, has also warned that regulation could hurt smaller competitors.
Other critics say the proposals are insufficiently ambitious. AI scientist David Krueger described the leaders’ call as “too little, too late,” while former Anthropic researcher Jacob Coxon said a coordinated international slowdown would be necessary if governments are to prevent a race in which each country accelerates because it fears falling
behind.
President Donald Trump has taken the opposing position. He has argued that the United States must win the AI race against China and has dismissed what he describes as excessive warnings about AI risks.
House Speaker Mike Johnson has said regulation matters but that Congress should not rush into action before it understands how a slowdown could work.
The China problem
The central geopolitical obstacle is that a unilateral U.S. slowdown could have the opposite effect from its intended goal.
If American companies reduce development while Chinese firms continue advancing, Washington could lose technological, economic and military advantages.
Amodei therefore argues for coordination that protects U.S. leadership, including tighter controls on the transfer of advanced chips and AI technology to authoritarian states. China’s response has so far been hostile.
A state-backed Global Times editorial
characterized Anthropic’s proposal as a Cold War-style attack on China’s AI industry.
That reaction suggests that a binding treaty to limit development speed would face substantial diplomatic and verification challenges.
A narrower agreement may be more attainable than a comprehensive cap on AI progress.
Governments could begin with shared definitions, incident reporting, model evaluations, restrictions on dangerous applications and communication channels for emergencies.
Such measures would not eliminate competition, but they could reduce the risk that safety failures are hidden or repeated across companies and borders.
What happens next
The debate is moving from abstract warnings to institutional design. The most consequential questions are now practical: Which systems count as “frontier” AI?
What capabilities trigger additional testing? Who appoints independent evaluators? Can evaluators inspect training processes without exposing trade secrets?
How should regulators handle open- source systems? And what penalties apply when a company fails to report a dangerous incident?
Obama and Harris have helped make those questions politically visible. Amodei’s proposal supplies one possible mechanism.
Industry support gives the idea momentum, but opposition from the White House, China and parts of Silicon Valley shows that agreement on the existence of risk is not agreement on the remedy.
The emerging middle ground is not a simple pause. It is a demand that the speed of deployment be tied to evidence that safety controls, independent testing and public accountability are keeping up.
Whether that becomes law, voluntary industry practice or another failed attempt at coordination will depend on decisions now being made by govts, companies and the public.
