AI safety or control? Yann LeCun challenges Anthropic over calls to slow down development
AI pioneer Yann LeCun has challenged calls for tighter restrictions on advanced artificial intelligence, questioning whether warnings about its dangers could serve corporate and political interests. Speaking in Seoul, the computer scientist defended open-source AI, criticised Anthropic’s restrictive approach to model releases and argued that safety concerns should be addressed through engineering rather than broad limits on access.
Yann LeCun aka the godfather of AI has challenged the growing push for tighter controls on advanced artificial intelligence, arguing that concerns about AI safety should not become a justification for slowing development or restricting public access to powerful models.
The Turing Award-winning computer scientist criticised Anthropic CEO Dario Amodei’s approach to AI safety during an interview at the Westin Chosun Seoul. LeCun questioned whether warnings about the technology’s potential dangers provide a sufficient basis for limiting access, suggesting that such concerns can also serve commercial and political interests.
His remarks highlight a widening debate within the AI industry. Where should developers draw the line between protecting the public from potential risks and preventing a small number of companies from controlling access to increasingly powerful technology?
AI safety or control?
LeCun sees Anthropic’s approach as excessively restrictive. While OpenAI CEO Sam Altman has advocated caution in releasing AI systems, according to LeCun’s account, Amodei has taken a more restrictive position on making powerful models publicly available.
The debate is not simply about how quickly AI should advance. It also concerns who gets to develop the technology, who can examine its limitations and who ultimately controls its deployment.
LeCun has argued that warnings about AI dangers can be used to influence government policy in ways that benefit established companies. He accused Anthropic of emphasising AI risks in a manner that could encourage governments to impose restrictions, according to South Korean publication Chosun.
Tech
He argues that limiting access to model weights, the numerical parameters that shape a trained model’s behaviour, could concentrate technical knowledge and influence among a small number of organisations.
Openly available models, by contrast, allow researchers and developers to study existing systems, adapt them for new applications and build on previous work. LeCun believes this wider participation is essential to ensuring that AI development does not become the preserve of a few powerful companies.
To illustrate his objection to restricting access to knowledge, he invoked historical attempts to limit the spread of information, comparing AI restrictions with the Ottoman Empire’s reported prohibition of printing in the 15th century. His broader argument is that limiting access to a transformative technology can impede the wider circulation of knowledge and expertise.
Why LeCun disputes AI danger warnings
LeCun’s opposition to broad restrictions is also rooted in his assessment of current AI capabilities. He has questioned whether existing large language models justify warnings about an existential threat to humanity, arguing that their limitations remain significant.
His position challenges the assumption that increasingly capable AI systems should automatically face increasingly restrictive release policies. In LeCun’s view, concerns about future risks must be weighed against what present-day systems can actually do and the technical measures available to manage their behaviour.
That does not mean every AI risk can be dismissed. Rather, LeCun argues that the response should be proportionate to the capabilities and limitations of the systems in question, instead of relying on worst-case scenarios to justify sweeping restrictions.
The disagreement reflects different approaches to uncertainty. Those advocating greater caution argue that developers may need to limit access when the potential consequences of releasing a powerful system are difficult to predict. LeCun, meanwhile, warns that broad restrictions could limit independent scrutiny and entrench the influence of the companies already developing the most advanced models.
The challenge for the industry is to determine how to manage risks without unnecessarily narrowing access to research, tools and technical expertise.
Who should control the future of AI?
Restricting access may help developers limit certain risks associated with releasing advanced systems. But extensive controls can also reduce opportunities for independent research, constrain smaller developers and give established companies greater influence over the direction of the technology.
Open-source development offers a different model, allowing a wider community to inspect, adapt and build upon AI systems. Yet wider access can also make it more difficult to control how a model is modified or deployed once its weights are publicly available.
Neither concern can be dismissed simply by invoking safety or openness. The challenge is to establish safeguards that address credible risks while preserving the benefits of research, competition and wider participation.
LeCun’s intervention brings that tension into sharper focus. By challenging Anthropic’s restrictive stance, he is questioning not only how developers should respond to AI risks but also who should have the authority to decide which systems can be accessed and by whom.
Ultimately, the debate is about more than whether AI development should slow down. It is about how to balance precaution with innovation, and public safety with the risk of concentrating technological power.
Handpicked stories, in your inbox
Global stories. Indian perspective. Zero noise.
No Spam. Unsubscribe Any Time.
First Published:Oct 14, 2026, 10:51:08 IST