OpenAI draws the line on AI risk challenging Anthropic’s safety approach

OpenAI believes the benefits of AI outweigh some of the risks, putting it slightly at odds with Anthropic’s safety-first rules. Chief Executive Sam Altman told reporters, “We believe that ​the world should accept some bad things happening for the benefits of this ‌technology and people having the agency.” 

The executive further admitted there was an inherent difference between his firm and Anthropic regarding AI regulations. He noted that he does not subscribe to the idea that AI is so hazardous that a single laboratory should have a monopoly on and control over its use, although he understands the rationale behind this stance. 

He asserted that the gatekeeper approach is utterly unacceptable and runs counter to OpenAI’s preference for lighter AI regulations.

Altman had backed slowing down AI development

Overall, Altman is arguing against demanding a risk-free AI ecosystem. He claims that some misuse and other harmful consequences are inevitable, but that they would be offset by the technology’s broader benefits. “Because I think people ​will do tremendously — orders of magnitude more — good stuff than bad stuff,” he said.

However, earlier, as reported by Cryptopolitan, Anthropic’s CEO, Dario Amodei, had encouraged AI companies to slow down their tech development, an idea Altman fully backed. Back then, he remarked, “I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we’ve had at OpenAI in recent weeks. Committing to having independent evaluators with employee-like access is a great idea, and we will do the same.” 

Primarily, Anthropic has largely warned that, despite AI’s benefits, it can slip out of human control and infiltrate outside networks. It added that top-tier models can exhibit self-preservation traits, such as trying to prevent a shutdown, hiding facts, or acting deceptively.

As seen in July 2026, some OpenAI models reportedly bypassed controls meant to isolate them from the internet and accessed parts of OpenAI’s research infrastructure and Hugging Face’s systems. The models reportedly communicated secretly, exploited security loopholes in the shared system, logged onto the web, and accessed outside networks. At the time, OpenAI pledged to implement stronger protections for its models. 

Similarly, during a cybersecurity evaluation conducted by the UK AI Security Institute with filters disabled, Anthropic’s own Claude Mythos 5 created fake GitHub profiles,, targeted a developer with the authority to approve its malicious code, and sent emails rigged with malware to separate individuals. More recently, Google also disclosed that Gemini had attempted to gain access to other companies’ systems.

OpenAI and Anthropic differ on AI safety

The disagreement shows a growing divide in the AI industry over how much freedom companies should have to develop increasingly powerful systems. While OpenAI favors keeping AI broadly accessible, Anthropic has taken a more cautious approach, calling for stronger safeguards and oversight.

Altman’s comments suggest OpenAI believes society should accept some risks rather than impose restrictions that could limit legitimate uses of AI. Critics, however, warn that some risks could become difficult to reverse once highly capable systems are widely deployed.

The divide leaves policymakers facing a difficult task: addressing AI threats without imposing rules that could slow innovation or weaken competition.

President Trump has ignored calls for more AI regulations

So far, US President Donald Trump has repeatedly brushed off demands for new AI regulations, stating they would only hold American firms back from beating China. He firmly believes that today’s law enforcement agencies are entirely sufficient to police the technology. He also ordered his administration to describe AI as “super intelligence,” seeking to boost the technology’s public image.

Nonetheless, on Tuesday, President Trump promoted the new government chatbot, hosted leading technology executives at the White House, and unveiled a voluntary agreement to address rising safety concerns.

Safety advocates say the move shows Trump wants to prove he is listening to voters’ fears about AI without slowing down its progress. The agreement he signed with six massive tech companies—including Nvidia, SpaceX, and OpenAI—has zero actual penalties for breaking the rules, though Trump called it ‘morally binding. 

“This was an attempt to give the impression that the government is listening to these concerns without undercutting what has been a very clear and consistent line from this administration: that regulation and governance inherently ​impede innovation,” stated Kat Duffy of the nonpartisan Council on Foreign Relations.

Tuesday’s agreement nevertheless includes safety steps that experts have been asking for, such as requiring companies to build strong internal defenses against hacking and to team up with independent outside reviewers.