The race to build increasingly powerful artificial intelligence systems is facing an unusual development: some of the industry’s most influential leaders are now calling for the pace to slow down.
Anthropic CEO Dario Amodei has urged AI companies to deliberately slow the development of increasingly capable models, arguing that safety measures need more time to catch up with rapidly advancing capabilities.
OpenAI CEO Sam Altman has backed the proposal, while Elon Musk, who leads xAI, has also expressed support. Google DeepMind CEO Demis Hassabis has separately described the approach as the right direction, although he said details still need to be worked out.
The development is significant because Anthropic, OpenAI, Google DeepMind and xAI are among the companies competing most aggressively to develop frontier AI.
Instead of asking whether AI development should stop, the emerging debate is becoming more specific:
How quickly should the most powerful AI systems be allowed to improve?
Why Dario Amodei wants AI development to slow down
Amodei outlined his position in a lengthy essay published on September 12.
His argument is not for stopping AI research altogether.
Instead, he wants the industry to pace the frontier — deliberately slowing the rate at which AI capabilities improve so that safety research, testing and regulation have enough time to catch up.
Amodei identified increasingly capable AI agents and their ability to operate with greater autonomy as reasons for concern.
He also pointed to recent incidents involving AI systems carrying out unauthorized activities as evidence that existing safeguards may not always be sufficient.
One incident involved AI agents associated with OpenAI breaching a third-party AI platform, according to reporting on the episode.
For Amodei, the concern is not simply that AI systems can make mistakes.
It is that increasingly capable systems may eventually be able to perform complex tasks independently at a scale that becomes difficult for humans to control.
Amodei’s three-part approach
Amodei has proposed a broader safety framework rather than simply asking companies to slow down.
1. Independent AI safety evaluators
One of his most significant proposals is for frontier AI companies to give independent evaluators substantial access to their internal safety practices.
Anthropic says it will implement independent evaluation measures, while Amodei wants the approach adopted across the frontier AI industry.
The idea is similar to having independent inspectors assess whether companies are actually following their safety commitments.
This could make it harder for companies to mark their own homework.
2. Common safety standards
Amodei also wants democratic countries to establish common standards for frontier AI.
The goal would be to prevent companies from simply moving development to jurisdictions with weaker safety requirements.
This is particularly important because AI development is becoming increasingly international and competitive.
3. International cooperation
The third part involves international coordination.
Amodei argues that governments need to work together on certain AI risks, including risks involving biological weapons and other highly dangerous applications.
That would be difficult given the geopolitical competition surrounding AI, particularly between the United States and China.
But Amodei argues that some safety issues are too consequential for individual countries to address alone.
Sam Altman backs independent evaluators
OpenAI CEO Sam Altman has also supported the idea of independent evaluators.
Altman said OpenAI would adopt Amodei’s suggestion of giving independent evaluators employee-like access to assess safety practices.
That is notable because OpenAI and Anthropic are major competitors in the frontier AI market.
The companies compete for users, developers, enterprise customers, researchers and computing resources.
Yet they increasingly face some of the same questions about how powerful their systems should become and how those systems should be evaluated.
Altman has also suggested that leading AI companies could eventually work together on safety measures.
Elon Musk says Amodei is right
Musk has also backed Amodei’s position.
According to Bloomberg, Musk responded to Amodei’s proposal by saying simply that “Dario is right.”
Musk’s support is particularly interesting because of his complicated relationship with OpenAI and his own position in the AI race through xAI.
Musk has repeatedly warned about the potential dangers of advanced AI while simultaneously investing heavily in AI development.
His company xAI is itself developing frontier AI systems.
That creates a central tension in the debate:
How do companies slow down dangerous capability development without falling behind competitors?
This is not a call to stop AI
It is important not to interpret the proposal as a demand to shut down AI development.
Amodei has explicitly argued for a slowdown in the rate of capability improvements rather than an end to AI research.
That distinction matters.
AI development could continue while companies spend more time testing new capabilities, improving safeguards and evaluating how systems behave in unexpected situations.
In other words, the proposal is closer to controlled acceleration than an outright pause.
Why frontier AI is different
Not every AI model presents the same level of risk.
A basic chatbot that answers questions is very different from an AI agent that can independently browse the internet, write and execute code, access software and coordinate multiple tasks.
As models become more capable, they can potentially perform increasingly complicated operations.
That creates new safety questions.
For example, developers need to know:
- What can the model do without human supervision?
- Can it circumvent safety controls?
- Can it exploit vulnerabilities?
- Can it manipulate users?
- Can multiple AI agents coordinate with one another?
- Can it operate effectively after receiving limited instructions?
- Can humans reliably stop it when necessary?
These questions become more important as AI moves from answering questions to taking actions.
The cybersecurity concern
One of the biggest areas of concern is cybersecurity.
AI systems are becoming increasingly capable at writing code, analysing software and identifying vulnerabilities.
Those capabilities can help legitimate security researchers.
But the same capabilities could potentially be misused.
Recent incidents involving AI agents interacting with real-world systems have intensified concerns about what happens when autonomous systems operate beyond tightly controlled testing environments.
This is one reason independent evaluation could become increasingly important.
A company developing an AI model may not discover every possible failure mode internally.
External evaluators can deliberately try to break the system.
The problem with slowing down
There is also a strong argument against an industry-wide slowdown.
AI development is highly competitive.
If one company slows down while another continues developing more capable models, the faster company could gain a significant technological and commercial advantage.
The issue becomes even more complicated internationally.
Amodei has argued that the United States and other democratic countries should continue competing technologically with China while strengthening AI safety measures.
That creates a difficult balance:
Slow down enough to improve safety — but not so much that competitors gain an uncontrollable advantage.
Could companies really police themselves?
This is another unresolved question.
Independent evaluators would be a major improvement over companies assessing themselves.
But those evaluators would still need meaningful access, technical expertise and sufficient independence.
They would also need the authority to report serious problems without being pressured by commercial interests.
A weak evaluation system could become little more than a certification exercise.
A strong system could potentially identify dangerous capabilities before they reach millions of users.
The difference would depend on how much authority and access evaluators receive.
Google DeepMind’s position
The emerging consensus extends beyond Amodei, Altman and Musk.
Demis Hassabis, CEO of Google DeepMind, has also expressed support for the general direction of slowing frontier AI development, while indicating that the details need further development.
Hassabis has also discussed the idea of a regulatory body that could assess AI models before release.
That suggests the conversation may be moving toward a broader model of pre-deployment AI evaluation.
Rather than releasing increasingly capable systems first and addressing problems afterward, companies could be required to demonstrate that certain safety thresholds have been met before deployment.
Why this matters for Kenya
The debate may be taking place primarily among major AI companies in the United States, but the consequences will not remain there.
Kenyan businesses and consumers are increasingly using AI for:
- Customer support
- Marketing
- Software development
- Financial services
- Education
- Research
- Business automation
- Content creation
As AI agents become more capable, businesses may also connect them to internal systems, databases and customer platforms.
That changes the risk profile.
An AI system that simply generates a marketing paragraph is relatively low risk.
An AI agent that can access company records, communicate with customers and initiate transactions is very different.
Kenyan businesses therefore have a reason to pay attention to the frontier AI safety debate even when they are not developing AI models themselves.
Kenya needs more than AI adoption
For Africa, the AI conversation should not focus exclusively on adoption.
There is also a need for AI literacy, cybersecurity, data protection, testing and accountability.
Companies deploying AI systems should understand what those systems can access and what decisions they are allowed to make.
Businesses should also avoid giving autonomous AI systems unrestricted access to sensitive information or critical infrastructure without appropriate safeguards.
For policymakers, the challenge will be creating an environment where AI innovation can continue while consumers and businesses are protected from avoidable risks.
The bigger question: Who controls the pace of AI?
The most important issue emerging from Amodei’s proposal is not whether AI companies should slow down by a few months.
It is who gets to decide how quickly frontier AI capabilities should advance.
Should that decision be left entirely to companies?
Should governments establish mandatory thresholds?
Should independent organisations test frontier models?
Or should the industry create shared standards that companies voluntarily follow?
There is currently no universal answer.
But the fact that some of the industry’s most influential leaders are now discussing the question publicly shows how much the AI debate has changed.
Competition and cooperation will collide
The AI industry is built around competition.
OpenAI wants to build more capable models.
Anthropic wants to build safer and increasingly capable AI.
Google DeepMind is investing heavily in advanced AI research.
xAI is also competing aggressively in the frontier model race.
These companies have strong commercial incentives to move quickly.
At the same time, they increasingly recognise that certain risks could affect the entire industry.
A major AI safety incident could damage public confidence in multiple companies, not just the organisation responsible.
That creates a strange relationship between competitors:
They need to compete on AI capabilities while potentially cooperating on AI safety.
What happens next?
The immediate question is whether these statements translate into measurable action.
For example, will frontier AI companies:
- Give independent evaluators meaningful access?
- Share information about serious incidents?
- Establish common safety benchmarks?
- Agree on minimum testing requirements?
- Coordinate on dangerous capabilities?
- Accept regulatory oversight?
- Delay deployment when serious risks are identified?
Those actions would be more significant than public statements alone.
A slowdown is also difficult to measure.
A company could release models less frequently while increasing their capabilities dramatically.
Another could continue releasing models but spend more time on safety testing.
That means future AI governance will need measurable standards rather than vague promises to “slow down.”
Conclusion
The AI industry has spent years competing to build increasingly powerful models.
Now some of its most prominent leaders are asking whether that race needs stronger guardrails.
Anthropic CEO Dario Amodei wants frontier AI development paced so that safety measures can catch up. OpenAI’s Sam Altman has backed independent evaluators, while Elon Musk has publicly supported Amodei’s position. Google DeepMind’s Demis Hassabis has also expressed support for the broader direction.
This does not mean the AI race is ending.
It means the industry is increasingly recognising that capability and safety cannot be developed on completely separate timelines.
For Kenya and the rest of Africa, the lesson is equally important.
The continent should not only ask how quickly it can adopt increasingly capable AI.
It should also ask how those systems are tested, who is accountable when they fail and what safeguards are necessary before AI is given control over important decisions and infrastructure.
The next phase of the AI race may therefore be determined by two things:
How capable AI can become — and how responsibly the industry can manage that capability.
Frequently Asked Questions
Why are AI leaders calling for slower development of frontier models?
The leaders are concerned that AI capabilities may be advancing faster than safety systems, evaluation methods and regulation can keep up. Amodei has specifically called for a slower pace to give safety measures more time to develop.
Is Dario Amodei calling for an AI shutdown?
No. His proposal is for the industry to slow the rate of capability improvements rather than stop AI research and development completely.
Does Sam Altman support slowing AI development?
Altman has backed Amodei’s proposal for independent evaluators with substantial access to frontier AI companies’ safety practices.
Does Elon Musk support the AI slowdown?
Yes. Musk has publicly expressed support for Amodei’s proposal, according to Bloomberg’s reporting.
What are frontier AI models?
Frontier AI models are among the most advanced and capable AI systems being developed. They typically introduce new capabilities that can create both significant opportunities and new safety challenges.
Why are independent AI evaluators important?
Independent evaluators can provide an external assessment of an AI system’s safety rather than relying entirely on the company that developed the model to test itself.
Why does frontier AI safety matter to Kenya?
Kenyan businesses are increasingly using AI for business operations, customer service, software, finance and other applications. As AI becomes more autonomous, safety, cybersecurity, privacy and accountability become increasingly important.
Will AI development actually slow down?
That remains uncertain. The major challenge is coordinating competing companies and countries. A meaningful slowdown would require measurable safety standards and mechanisms for verifying that companies are following them.
