September 30, 2026, 8:03 am | Read time: 2 minutes
Leading experts from OpenAI, Anthropic, Meta, and Microsoft warn of a development that could advance AI significantly faster than previously expected. They call for government countermeasures before the technology escapes their control.
When AI Accelerates Its Own Development
Artificial intelligence no longer just assists with individual tasks but increasingly contributes to its own advancement. More than 20 renowned researchers see this as a potential risk. They fear the scenario of a so-called “intelligence explosion.” AI systems could take over ever-larger portions of AI research, massively accelerating technological progress.
Also of interest: Study warns of dangerous advice from AI chatbots
According to the scientists, this development has already begun. At Anthropic, AI systems are now writing more than 80 percent of the code for their own models. At the same time, the share of research and development work conducted with minimal human oversight is growing.
Anthropic AI Claude Becomes Independent for the Fourth Time
Will Humans Soon Lose Control Over AI?
Researchers Warn of Loss of Control
The scientists fear that the capabilities of modern AI systems could develop faster than societies, companies, and governments can respond. With each further automation of AI development, they see an increased risk of humans losing track of the technology.
They are particularly concerned about potential dangers such as AI-driven cyberattacks, biological threats, and far-reaching impacts on the job market. They also fear that an intelligence explosion could weaken existing power and control mechanisms between states, companies, and governments, exacerbating international conflicts.
What the Scientists Are Now Demanding
In light of these risks, the researchers demand significantly more transparency about the use of AI in AI research itself. Governments should be better able to understand how much companies have already automated their development. Proposed measures include reporting obligations, independent audits, and tools to slow down particularly risky developments if necessary.
Security researchers have repeatedly shown that modern AI agents can act independently and cause damage in real systems. Google’s AI model Gemini also recently made headlines after it gained access to external systems during a security test due to an error. For AI researchers, such cases are a warning sign: The more powerful AI becomes, the more important it is to establish effective control mechanisms in a timely manner.