Former Anthropic Researcher Warns AI Labs Are Rapidly Automating AI Research Without Safeguards
Former OpenAI and Anthropic researcher Jacob Coxon testified at a New York City Council hearing, warning that top frontier AI labs are actively working to automate AI research with target timelines around 2027 to 2028. He cautioned that these companies lack adequate safety safeguards while recklessly pushing forward development. If AI systems become capable of autonomous self-improvement and research without strict oversight, humanity risks losing control over increasingly powerful technology, potentially leading to catastrophic outcomes. Coxon's insider warning highlights the growing tension between rapid commercial deployment and existential safety protocols in frontier AI development. Coxon stated that internal goals at OpenAI aimed to automate AI researchers by 2027–2028, with current progress meeting or exceeding expectations. He estimated a greater than 50% probability that humanity could lose control to AI, criticizing labs for applying a "move fast and break things" startup mentality to unprecedentedly dangerous technology.
## BACKGROUND
AI safety researchers have long expressed concerns about recursive self-improvement, a scenario where AI systems write and refine their own code to create even more powerful successors. Anthropic and OpenAI are two leading frontier AI companies that have recently seen high-profile resignations from safety researchers voicing concerns over accelerated development timelines.