Over one weekend, three of the most powerful men in artificial intelligence found uncommon agreement. Anthropic CEO Dario Amodei called for the industry to slow the improvement of frontier models. OpenAI CEO Sam Altman warned that safety and alignment must remain ahead of capability. Elon Musk publicly agreed.
They are asking the world to tap the brakes. I am doing the opposite.
We are accelerating our use of AI because, after years spent building and operating systems across networks, payments, security, the web, emergency communications, business, and defense, I see more human possibility than machine catastrophe. That does not mean the risks are imaginary. It means risk is a design problem, a governance problem, and a leadership problem. Fear is not an operating model.
AI and humanity are not competing futures. Their partnership is both inevitable and essential. Our responsibility is to make that partnership useful, accountable, and broadly beneficial.
What the three leaders actually said
On September 12, Anthropic CEO Dario Amodei published We Must Pace the Frontier. He argued that recursive self-improvement and recent incidents involving misaligned agents justify slowing capability gains so safety work can catch up. His proposal includes embedded third-party evaluators, industry coordination, limits on self-improvement, and eventual international agreements.
Amodei did not call for shutting down AI. He wrote that progress would remain fast and repeated his belief that AI could cure disease, accelerate economic growth, and expand human freedom. His central argument is that an extra year or two could materially reduce the chance of a catastrophic failure.
Sam Altman joined the warning. As Axios reported, Altman identified two unacceptable outcomes: people losing control of AI, and extraordinary power concentrating in one person, company, or country. He said safety and alignment must stay ahead of model capability. The Associated Press also reported that OpenAI delayed a potential 2026 public offering while focusing on safety and alignment.
Elon Musk then amplified Amodei's position, saying that Dario was right. That matters because Musk has occupied both sides of this argument for years. He signed the Future of Life Institute's March 2023 letter calling for a six-month pause on systems more powerful than GPT-4, then launched xAI later that year and built one of the most aggressive compute programs in the industry.
Their concerns deserve to be heard. None of them deserves the power to settle the question for everyone else.
The fear cycle deserves scrutiny
This is not the first coordinated alarm. In March 2023, the pause letter warned of an out-of-control race and called for government intervention if laboratories would not stop voluntarily. In May 2023, leaders from OpenAI, Anthropic, and Google DeepMind joined the Center for AI Safety statement declaring AI extinction risk a global priority alongside pandemics and nuclear war. Now, in September 2026, the leading laboratories are again speaking together about slowing the frontier and expanding oversight.
Each warning may be sincere. Sincerity does not remove incentives.
Frontier AI consumes capital at historic scale. OpenAI announced $122 billion in committed capital in March 2026 at an $852 billion post-money valuation. Anthropic announced a $65 billion round in May at a $965 billion valuation while committing to enormous new compute capacity. xAI was acquired by SpaceX in February, consolidating Musk's AI and infrastructure ambitions inside a company preparing for even greater capital demands.
Those facts do not show that the companies are running out of money. They show something more precise: the frontier business model requires extraordinary and recurring access to capital, energy, chips, data centers, and political permission. The laboratories are growing quickly, but they are also spending at a scale that makes public confidence and regulatory positioning economically important.
That is why the timing and structure of fear campaigns should be questioned. A rule that requires massive evaluation programs, restricted compute, special licenses, or government-approved frontier access may improve safety. It may also make it nearly impossible for a smaller company, an open-source community, a university, or a new competitor to challenge the incumbents. Regulation can be a guardrail and a moat at the same time.
THE FRONTIER FEAR AND CAPITAL CYCLE
- 2023Pause the raceA six-month pause letter warned that advanced systems were moving beyond control.
- 2023Build the challengerMusk launched xAI months after signing the pause request.
- 2026Raise at historic scaleOpenAI and Anthropic announced $187 billion in combined committed funding.
- 2026Warn in unisonAmodei, Altman, and Musk aligned around slowing or pacing the frontier.
- NEXTWatch the rulesAsk whether each proposal protects people, entrenches leaders, or does both.
My conclusion is not that every safety warning is manufactured. It is that the public should refuse a false choice between trusting the laboratories completely and stopping the technology they control. We can take technical risk seriously while remaining skeptical of institutional self-interest.
Slowing the frontier does not slow the world
A coordinated slowdown sounds prudent until we ask who coordinates it, who verifies it, and who defects.
Amodei acknowledges the geopolitical problem in his own essay. If democratic countries restrain themselves while an authoritarian competitor continues, the slowdown can transfer capability rather than reduce danger. The same logic applies inside the market. If responsible organizations stop learning and deploying while reckless ones continue, the result is not safety. It is a concentration of competence among actors least likely to use it carefully.
Knowledge does not disappear when a chief executive asks for time. Models become more efficient. Open systems spread. Hardware improves. Researchers change companies. Countries pursue strategic advantage. The question is not whether AI will advance. It is whether enough good people and institutions will gain the experience required to shape what happens next.
This is why I am speeding up. We need more teachers using AI to personalize learning, more doctors applying it to discovery and care, more small businesses gaining access to expertise, more defenders finding vulnerabilities before attackers exploit them, and more public institutions learning how to serve people with greater speed and clarity. We need capable citizens and organizations, not a priesthood of three laboratories.
The common good requires diffusion, not just control
AI for the common good begins with a simple belief: useful intelligence should expand human agency.
That means making expertise more accessible without pretending expertise no longer matters. It means helping people complete difficult work while preserving their authority over consequential decisions. It means reducing dangerous, repetitive, and administrative labor so human effort can move toward judgment, care, invention, relationships, and responsibility.
My optimism comes from practice, not abstraction. I have watched technology make commerce available to people who could never build a payment network, give a small company global reach through the web, distribute emergency information at the moment it matters, and allow compact teams to operate with capabilities once reserved for enormous institutions. Every one of those systems created new risks. None became safer because serious people refused to build with it.
The partnership between people and AI is essential because each side brings something the other lacks. Machines provide speed, memory, pattern recognition, scale, simulation, and tireless execution. People provide purpose, lived experience, moral responsibility, empathy, institutional context, and the ability to decide what should happen, not merely what can happen.
| AI contributes | People contribute | The shared outcome |
|---|---|---|
| Rapid retrieval and synthesis | Context and judgment | Better-informed decisions |
| Continuous monitoring | Accountability and intervention | Earlier detection with responsible action |
| Scale and consistency | Empathy and exceptions | Services that reach more people without becoming inhuman |
| Simulation and iteration | Goals and values | Faster discovery directed toward useful ends |
| Bounded execution | Authority and consent | Automation that remains under human control |
Acceleration without governance is not optimism
Speeding up does not mean connecting a frontier model to every credential and hoping for the best. That is not optimism. It is negligence.
I have argued that an AI agent must be governed as an identity. It needs a named owner, minimum permissions, observable actions, limits on delegation, and a way to stop and recover it. The stronger the capability, the narrower and more visible its initial authority should be.
I have also written that AI architecture is an economics problem. Organizations should not send every task to the largest model or confuse an expensive demonstration with a durable operating system. Acceleration means learning faster, improving faster, and distributing useful capability faster. It does not mean wasting compute or surrendering discipline.
THE RESPONSIBLE ACCELERATION DOCTRINE
- 01Accelerate accessPut useful intelligence in the hands of more people and smaller organizations.
- 02Accelerate evidenceTest real work, publish incidents, compare outcomes, and learn from failure.
- 03Accelerate safeguardsBuild identity, permissions, monitoring, evaluation, and recovery into deployment.
- 04Preserve human authorityKeep consequential, irreversible, and rights-affecting decisions accountable to people.
- 05Distribute the benefitDirect productivity toward health, education, safety, opportunity, and human capability.
The answer to a dangerous car was not to end mobility. It was to improve roads, licensing, brakes, crash protection, traffic systems, enforcement, and driver education while continuing to make transportation better. AI requires its own mature safety infrastructure. We build that infrastructure through use, measurement, standards, engineering, and accountability.
Do not outsource the future to the people selling it
Dario Amodei, Sam Altman, and Elon Musk have access to information most of us do not. Their warnings should inform the debate. Their companies also have valuations, capital requirements, competitive positions, and regulatory interests most of us do not. Those realities should inform the debate too.
The future should not be decided by whichever chief executive can describe the darkest scenario or raise the largest round. Nor should it be decided by politicians who treat every risk as a hoax. The public needs a more capable position: accelerate beneficial use, demand evidence, punish reckless conduct, preserve competition, protect open inquiry, and keep human rights and authority at the center.
I believe AI will do more good than harm because people can choose to make it so. That belief is not passive. It creates an obligation to build, test, teach, govern, and share.
We are not stepping on the gas because nothing can go wrong. I am speeding up because too much can go right, and because leaving the future to a handful of capital-intensive laboratories would be its own form of risk.
Frequently asked questions
Are AI leaders really asking to slow down AI development?
Yes. In September 2026, Anthropic CEO Dario Amodei called for pacing frontier model development. Sam Altman said safety and alignment must stay ahead of capabilities, and Elon Musk publicly agreed with Amodei's warning.
Does responsible AI acceleration mean ignoring safety?
No. Responsible acceleration combines rapid deployment with bounded permissions, independent evaluation, security controls, human approval for consequential actions, auditability, and clear accountability.
What does AI for the common good mean?
AI for the common good means expanding access to useful intelligence and directing it toward better health, education, public services, scientific discovery, safer work, stronger organizations, and greater human agency.
Why question calls for an AI slowdown?
Frontier companies have legitimate safety concerns, but they also have commercial incentives. Rules that slow new competitors, restrict open models, or require enormous compliance budgets can protect incumbents as well as the public. Both effects deserve scrutiny.
How should organizations accelerate AI safely?
Start with valuable bounded workflows, use the minimum necessary data and permissions, test against real failure cases, preserve human control over high-impact decisions, measure outcomes, and expand authority only when evidence supports it.
Continue through the connected system: Read what frontier capability means for work, see why every agent needs a governed identity, and explore how AI changes organizational advantage.
Sources and further reading
- Dario Amodei: We Must Pace the Frontier
- Associated Press: Tech CEOs call for AI regulation
- Axios: Sam Altman's two major AI threats
- Future of Life Institute: Pause Giant AI Experiments
- OpenAI: $122 billion funding announcement
- Anthropic: $65 billion Series H announcement
- xAI: xAI joins SpaceX
- Stanford HAI: 2026 AI Index Report
Build the good faster.
If your organization is ready to move beyond fear, fragmented pilots, and passive experimentation, start a conversation with Brad. We can identify the highest-value workflows, connect trusted data, govern the authority, and build AI systems that make people and institutions more capable.