Two former Anthropic researchers have raised $200 million just weeks after departing the company. Behnam Neyshabur and Harsh Mehta left in December 2025, following the launch of Claude Opus 4.5, to build a startup focused on AI systems that can create and improve other AI models autonomously.
The round's investors and valuation were not disclosed, according to TechFunding News. The funding arrives unusually fast after the founders' exit, signaling strong investor appetite for the concept of self-improving artificial intelligence.
Their approach targets a growing market for AI that can automate its own development, potentially reducing human oversight in model training and iteration. This places them in competition with labs like Google DeepMind and OpenAI, which are also exploring recursive self-improvement techniques. The startup's early traction suggests investors are betting on a paradigm shift in how AI systems are built.
This development highlights a trend toward autonomous AI development tools, which could accelerate progress but also raise safety concerns. The departure of key researchers from established labs to launch competing ventures underscores the intense demand for talent and novel approaches in frontier AI. Observers will watch whether self-improving AI can deliver on its promise without introducing uncontrollable feedback loops.
No additional details about the founders' academic background or specific technical plans were provided in the report. The startup's name and product roadmap remain undisclosed.
Counter-argument: Some experts warn that AI systems capable of improving themselves could pose existential risks if safety measures are not rigorously implemented, and the rapid exit from Anthropic may raise questions about adherence to responsible AI development practices.