AI Leaders Are Getting More Candid about the Risks

“Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”

In May 2023, leaders from the world’s most influential AI companies signed that statement. Among them were Sam Altman of OpenAI, Amodei of Anthropic, and Demis Hassabis of Google DeepMind. [internatio...report.org], [edexlive.com]

It remains one of the most extraordinary statements in the history of technology. Inventors are usually the biggest champions of their inventions. Yet the leaders building the world’s most advanced AI systems are increasingly talking about risk, uncertainty, and the possibility that society may not be fully prepared for what comes next. [internatio...report.org], [cnbc.com]

What makes September 2026 different is not that AI critics are worried. Critics have warned about AI risks for years. What’s different is that researchers inside the leading AI labs are publicly debating whether AI capabilities are advancing faster than humanity’s ability to understand, govern, and control them. Former and current researchers from Anthropic, OpenAI, and Google DeepMind have publicly raised concerns about alignment, oversight, and the pace of development. [cnbc.com], [timesnownews.com], [moneycontrol.com]

The conversation is no longer confined to academic papers, conferences, or private discussions. It’s happening in public, and it is increasingly being driven by the people closest to the technology. [cnbc.com], [timesnownews.com]

Why September 2026 Feels Different

The AI safety debate is not new. Concerns about superintelligence, alignment, and loss of control have circulated among researchers for more than a decade. What is new is the willingness of researchers at frontier AI laboratories to speak openly about those concerns. [edexlive.com], [theguardian.com]

Over the past several weeks, former and current employees of Anthropic, OpenAI, and Google DeepMind have publicly argued that increasingly powerful AI systems may be approaching capabilities that safety research and governance institutions are not yet equipped to manage. Whether those concerns ultimately prove correct remains an open question. But the fact that they are now being expressed publicly by insiders represents a meaningful shift in the AI conversation. [cnbc.com], [timesnownews.com], [moneycontrol.com]

What Anthropic Is Saying

If one company captures the mood of September 2026, it is Anthropic.

Since its founding, the company has positioned itself around a central idea: advances in AI capabilities should be matched by advances in AI safety. CEO Amodei has consistently argued that advanced AI could drive breakthroughs in medicine, science, and economic growth while simultaneously creating unprecedented risks if researchers fail to solve alignment and control challenges. [edexlive.com], [internatio...report.org]

That tension came into sharp focus this month when former Anthropic researcher Jacob Coxon resigned publicly and accused leading AI companies of “gambling with our lives.” His departure triggered a remarkable public discussion among current and former AI researchers. Several researchers associated with Anthropic openly supported parts of his warning and argued that society should take the possibility of severe AI-related risks more seriously. [ca.news.yahoo.com], [cnbc.com]

The debate centered on several issues:

• Whether future AI systems could become difficult to control. [edexlive.com], [timesnownews.com]

• Whether alignment research is progressing quickly enough. [cnbc.com], [moneycontrol.com]

• Whether AI systems could eventually contribute to the development of more capable successor systems through recursive self-improvement. [cnbc.com], [theguardian.com]

• Whether safety testing should determine when development slows or pauses. [ca.news.yahoo.com], [cnbc.com]

These concerns are reflected in Anthropic’s broader strategy. The company has invested heavily in alignment research, catastrophic-risk evaluations, and its Responsible Scaling Policy, a framework designed to increase safety requirements as AI systems become more capable. [cnbc.com], [internatio...report.org]

Anthropic’s message is not that AI development should stop. Rather, it argues that capability improvements and safety improvements must advance together because society should not assume that control problems have already been solved simply because today’s systems perform well. [edexlive.com], [internatio...report.org]

What OpenAI Is Saying

While Anthropic has been at the center of this month’s controversy, OpenAI remains the company most closely associated with the AI revolution.

CEO Sam Altman continues to argue that AI could become one of the most beneficial technologies ever developed. OpenAI regularly points to the technology’s potential to accelerate scientific research, education, healthcare, and economic productivity. At the same time, the company has long acknowledged the risks associated with increasingly capable systems. [internatio...report.org], [cryptobriefing.com]

During testimony before the U.S. Senate on May 16, 2023, Altman warned lawmakers:

“If this technology goes wrong, it can go quite wrong.” [internatio...report.org]

That statement has become increasingly relevant as AI capabilities continue to advance. OpenAI’s strategy remains centered on “iterative deployment,” the belief that society learns how to govern advanced AI by deploying systems gradually, studying their behavior, and improving safeguards over time. [edexlive.com]

More recently, OpenAI has emphasized preparedness frameworks, external evaluations, and frontier-model testing. The company has repeatedly argued that increasingly powerful systems should undergo extensive safety evaluation before release. [internatio...report.org], [edexlive.com]

Perhaps most notable is that concerns are no longer coming only from critics outside the company. Following the controversy surrounding Anthropic, members of OpenAI’s technical staff publicly supported calls for slowing AI development if safety research does not keep pace with capability improvements. Some researchers argued that stronger safeguards may be necessary as systems become increasingly capable. [cnbc.com]

The OpenAI position has evolved into something more nuanced than simple optimism. The company still believes AI should continue advancing, but it increasingly emphasizes preparedness, testing, and governance alongside progress. [edexlive.com], [internatio...report.org]

What Google DeepMind Is Saying

If Anthropic has become the most vocal about safety and OpenAI remains focused on deployment and preparedness, Google DeepMind offers the broadest long-term perspective.

CEO Demis Hassabis often frames AI as a scientific tool rather than simply a commercial technology. He regularly discusses its potential to accelerate breakthroughs in biology, medicine, mathematics, and fundamental scientific research. At the same time, DeepMind has consistently emphasized governance, international cooperation, and oversight. [internatio...report.org], [cryptobriefing.com]

Like Altman and Amodei, Hassabis signed the May 2023 AI safety statement. Since then, he has repeatedly advocated for institutions capable of managing systems that may become significantly more capable than today’s models. [internatio...report.org], [edexlive.com]

Those concerns became more visible this month when former Google DeepMind AGI safety researcher Josh Engels announced his departure from the company. Engels argued that researchers still lack reliable methods for ensuring that increasingly capable AI systems remain aligned with human intentions as their capabilities improve. [timesnownews.com], [moneycontrol.com]

His concerns echo broader discussions occurring throughout the industry. How should powerful AI systems be evaluated? How can alignment be verified? What oversight mechanisms should exist before more capable systems are deployed? And what role should governments play in supervising increasingly powerful technologies? [timesnownews.com], [moneycontrol.com], [internatio...report.org]

DeepMind’s central message is clear: if AI becomes one of humanity’s most important inventions, it will require institutions, standards, and governance frameworks equal to its importance. [internatio...report.org], [cryptobriefing.com]

If They’re So Concerned, Why Keep Building AI?

An obvious question follows: if the people building advanced AI are concerned about its risks, why not stop?

Their answer appears to rest on three arguments. First, they believe the potential benefits are enormous. Advanced AI could accelerate medical breakthroughs, scientific research, productivity growth, and economic development. Second, many researchers argue that AI safety cannot be studied entirely in theory. Understanding how highly capable systems behave requires observing those systems directly. Third, there is intense competitive pressure. No company wants to fall substantially behind in a field that could reshape industries and economies. [edexlive.com], [internatio...report.org]

This tension sits at the heart of the modern AI debate. The companies see extraordinary reasons to continue developing AI, but they also see growing reasons for caution. [edexlive.com], [cnbc.com]

What All Three Labs Agree On

Despite competing aggressively for talent, funding, computing resources, and market share, the three leading AI labs increasingly appear to share several core beliefs:

• AI capabilities are advancing rapidly and may continue accelerating. [cnbc.com], [internatio...report.org]

• Advanced AI could generate enormous benefits for medicine, science, education, and economic growth. [edexlive.com], [internatio...report.org]

• Future AI systems are likely to become substantially more capable than today’s models. [internatio...report.org], [theguardian.com]

• Safety research must advance alongside capabilities. [cnbc.com], [internatio...report.org]

• Increasingly powerful systems should undergo extensive testing and evaluation before deployment. [internatio...report.org], [cryptobriefing.com]

• Governments and institutions will need to play a larger role in oversight. [internatio...report.org], [cryptobriefing.com]

• Alignment, cybersecurity, misuse, and governance are becoming central concerns rather than secondary ones. [mitsloan.mit.edu], [tech-insider.org]

In other words, the debate inside the leading AI labs is no longer about whether AI matters. The debate is increasingly about how to manage its success.

The Real Story

The most important fact about today’s AI discussion is not that critics are worried. Critics are always worried.

The more significant development is that many of the researchers and executives building advanced AI systems are publicly discussing the same concerns. The leaders of OpenAI, Anthropic, and Google DeepMind remain optimistic about AI’s future. They continue to describe a world in which AI transforms science, medicine, education, and economic growth. [edexlive.com], [internatio...report.org]

But September 2026 revealed a shift. Questions once largely confined to internal safety teams are now being debated in public. Sam Altman, Amodei, and Demis Hassabis disagree on many details, yet they increasingly appear to agree that as AI capabilities accelerate, questions of alignment, testing, cybersecurity, safety, and governance are becoming more urgent. [cnbc.com], [timesnownews.com], [moneycontrol.com]

Investors, policymakers, business leaders, and the public do not need to know exactly what AI will look like ten years from now. But they should pay attention when the people closest to the technology begin openly debating whether society is ready for what comes next. [cnbc.com], [timesnownews.com], [moneycontrol.com]