For the past few years, the race between OpenAI and Anthropic has been about one big idea: building AI systems that can do far more than answer questions, write code or generate images. OpenAI has openly described its mission as ensuring that artificial general intelligence, or AGI, benefits all of humanity. Anthropic, meanwhile, has said that it is preparing for a future with increasingly powerful AI systems that could eventually match or exceed human experts across many fields.Now, the conversation inside both companies is changing.The two AI labs are still building increasingly capable systems. But their leaders are also warning that AI development may be moving so quickly that safety measures could struggle to keep up. Earlier this month, Anthropic CEO Dario Amodei called for the industry to slow the pace of AI development. OpenAI CEO Sam Altman agreed that frontier AI development needs to be paced and that stronger safety measures are necessary. That creates a striking shift in the AI race. The same companies that spent years trying to make machines more capable are now spending more time asking what happens when those machines become capable enough to act with much greater independence.
OpenAI’s goal was always AGI. The safety problem came with it
OpenAI was founded in 2015 with a mission centred on advanced AI. Its stated goal was to develop artificial intelligence in a way that would benefit humanity. The company’s later charter defined AGI as highly autonomous systems that outperform humans at most economically valuable work.OpenAI has not abandoned that goal. In 2026, it continues to describe AGI as central to its mission. Its current public policy agenda says advanced AI can create risks involving cybersecurity, chemical and biological threats and other areas, while also arguing that AI could bring major economic and social benefits.But OpenAI’s own recent research shows why the safety conversation has become harder.This month, the ChatGPT-maker said its upcoming Astra model had reached its “Critical” threshold for cybersecurity capability. OpenAI said that, with the right tools and access, Astra could find previously unknown security flaws and develop ways to exploit them across well-protected systems without a person guiding every step. The company said this required stronger safeguards before deployment.OpenAI has also started talking more openly about unexpected behaviour from its models. The company recently said that it will regularly disclose cases of AI systems showing unexpected or unauthorised behaviour.So the concern is no longer only about an imaginary future where AI becomes smarter than humans. The labs are increasingly dealing with systems that can already act more independently, use tools, operate computers and perform complex tasks.
Anthropic built its identity around safer powerful AI
Anthropic, on the other hand, was created with a different emphasis. The company says its mission is to help the world safely transition through transformative AI. Its Claude Constitution says Anthropic believes AI could be one of the most world-changing and potentially dangerous technologies in human history, while also arguing that a safety-focused company should remain at the frontier if powerful AI is coming anyway.That safety focus has produced formal rules around how Anthropic develops increasingly capable models. In 2023, the Claude-maker introduced its Responsible Scaling Policy, designed to address catastrophic risks from increasingly powerful AI systems. It created different safety levels and said stronger safeguards would be required as models crossed specific capability thresholds.Anthropic has since expanded that work. Its current policy says frontier AI could bring major benefits in areas such as science, healthcare and education, but also presents new risks that require safeguards.The company has also created the Anthropic Institute, which it says will study the effects of increasingly powerful AI on society. Anthropic said in March that its models had progressed from early commercial systems to models that could find serious cybersecurity vulnerabilities, perform real-world work and begin accelerating AI development itself.Then came Amodei’s latest warning.
Dario Amodei says AI development may be moving too fast
Earlier this month, Anthropic CEO Dario Amodei called for the AI industry to slow down the pace at which frontier models are being improved. His warning came after a series of incidents involving AI agents, including the OpenAI-Hugging Face incident that raised questions about what autonomous AI systems could do when given tools and access.Amodei said:“We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.”He also warned that the consequences could become much larger as AI systems improve.“Given the accelerating rate of AI capability development, it’s my worry that in 6–12 months such a swarm could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage), and that the scale of damage would continue to increase from there if AI becomes more powerful without the necessary guardrails,”Those are Amodei’s projections, not an established prediction about what AI will actually do. But the significance is that the warning is coming from the head of a company actively building frontier AI.
And then Sam Altman agreed
Dario Amodei’s call for slower development was notable because it came from a direct competitor of OpenAI. But OpenAI CEO Sam Altman soon said he agreed with the central argument.Sam Altman said the frontier needs to be paced and that OpenAI would adopt Amodei’s proposal for independent evaluators with employee-like access. A few days later, Altman also described two ways AI progress could go badly: humans losing control of AI and powerful AI becoming concentrated in the hands of one person, company or country.“This is unacceptable; we are unapologetically on Team Humanity, and AI must always serve people,” Altman said.That is a notable change in the public conversation.OpenAI is not saying it wants to stop developing more capable AI. Anthropic is not saying it wants to abandon frontier models either. Instead, both are arguing that capability growth and safety work need to move together.
The race is now about more than human-level intelligence
There is another reason the language has changed. The target is no longer simply an AI that can answer questions as well as a person. OpenAI says its latest systems can operate computers and graphical interfaces, collaborate with people and other AI systems, and carry out research projects.Anthropic has similarly described the next generation of powerful AI as systems that could navigate digital interfaces, browse the internet, reason through complex tasks for days or weeks and interact with physical equipment. The company has said such systems could emerge as soon as late 2026 or 2027, based on its research trajectory.That is why the debate around AGI is changing. The question is no longer only “Can AI become as intelligent as humans?” It is increasingly “What happens when AI can use that intelligence independently, for long periods, with access to the tools and systems around us?”And that distinction matters.
