PATTAYA, Thailand – Artificial intelligence development may be approaching a critical turning point, according to U.S.-based AI startup Anthropic, which has warned that future systems could soon be capable of creating, training, and improving themselves without direct human involvement.
In a paper published Thursday, Marina Favaro, head of Anthropic’s research institute, and co-founder Jack Clark cautioned that AI development is accelerating at a pace that could eventually outstrip human oversight. They argued that governments and industry leaders should consider slowing the speed of development to allow more time to assess the risks and implications of increasingly autonomous systems.
According to the report, today’s AI agents are already capable of writing and executing code, delegating tasks to other AI systems, and managing workflows that previously required hours of human effort. Anthropic says the company is gradually transferring more of its own development work to AI-assisted systems, significantly increasing productivity and development speed.
The authors suggested that if current trends continue and sufficient computing resources remain available, future AI models could eventually design and develop their own successors with minimal or no human intervention.
Concerns over self-improving AI have been growing across the industry. In late 2025, OpenAI disclosed that it was researching ways to safely develop and deploy highly capable systems, including models able to improve themselves over time. The company stated that such systems must remain aligned with human intentions, avoid harmful behavior, remain auditable, and operate within accepted social values.
Anthropic estimates that AI capabilities are now doubling roughly every four months, compared with earlier expectations of seven-month cycles. The company revealed that its Claude model currently writes around 80 percent of the code that is eventually integrated into its internal codebase.
While Anthropic stressed that fully autonomous self-improving AI is not imminent, the company warned that the transition could arrive sooner than many institutions expect. Once machine-generated code becomes comparable in quality to human-written software, human developers may shift primarily into supervisory roles. If humans can no longer review code as quickly as AI can generate it, human oversight itself could become the main bottleneck in future AI development.
The company acknowledged the dilemma facing policymakers. Slowing development could provide valuable time to evaluate safety risks, but it could also allow less responsible actors to catch up technologically. Without international coordination, governments and organizations may face increasingly difficult choices shaped by competition, economic pressures, and national security concerns.
Anthropic recently demonstrated its cautious approach to AI safety by withholding public release of an advanced model known as Claude Mythos after internal assessments found it could easily identify and create software vulnerabilities, raising cybersecurity concerns. The warning comes as AI agents are rapidly gaining traction in the cryptocurrency sector. Industry leaders increasingly view autonomous systems as a major driver of adoption and transaction growth. Circle CEO Jeremy Allaire predicted earlier this year that billions of AI agents could be operating on behalf of users within the next five years.
Evidence of that trend is already emerging. A May 2026 report by crypto investment firm Keyrock found that AI-powered payment agents have moved from theory to real-world deployment over the past year. According to the report, autonomous agents have already processed more than $73 million in payments across 176 million transactions. The findings suggest that artificial intelligence is no longer confined to research labs. Increasingly, autonomous systems are beginning to participate directly in the digital economy, managing transactions, interacting with financial networks, and potentially transforming industries ranging from software development to cryptocurrency and beyond.
