Artificial intelligence systems could automate most AI research and development (R&D) work within the next few years, potentially triggering an “intelligence explosion” in which years of AI progress are compressed into months or less, according to a new Cambridge working paper co-authored by Geoffrey Hinton, Yoshua Bengio and more than 20 other AI researchers and policy experts.
The paper, “What if automating AI R&D triggers an intelligence explosion?”, says preliminary evidence suggests that increasing automation of AI R&D could create a feedback loop in which AI systems help improve AI research, producing more capable systems that can accelerate research further.
“As more of the AI research and development (R&D) pipeline is automated, could AI progress radically accelerate in an “intelligence explosion,” where years of advances are compressed into months or less? Preliminary evidence suggests that it could,” the researchers noted.
According to the researchers, AI systems are already taking on major parts of the AI R&D pipeline. Anthropic reported that the share of the approved code written by AI systems increased from low single digits in January 2025 to more than 80 percent by May 2026. The proportion of R&D work autonomously completed with only high-level human supervision also rose from 1 percent in March to 26 percent in August 2026.
The paper further highlighted that OpenAI has reported that “AI assistance is used in practically all parts of the company across technical and non-technical teams with code-executing agents used in training, evaluating, and securing future models.”
Similarly Google has also mentioned that “AI is used in almost all work that involves writing code or configuration, technical design, research ideation, to different degrees depending on the task.”
The paper said the best AI systems can now complete AI R&D tasks that take human experts hours to days, compared with seconds-long tasks in 2023.
“Tentative extrapolations” of recent trends suggests that months-long AI R&D projects could be automated by mid 2028, the paper mentioned. It added that even the possibility of full automation within this timeframe should be taken seriously.
The proposed intelligence-explosion mechanism involves two stages. First, increasingly capable AI systems expand the effective R&D workforce by performing research tasks faster and better. Second, that expanded AI workforce develops even more capable systems, creating a recursive feedback loop.
The researchers estimate that if AI systems reached expert level AI R&D capabilities at costs comparable to today’s systems, the compute available to a single frontier AI developer could potentially support an AI workforce equivalent to at least millions of top human researchers.
However, the paper stressed that the size and duration of any resulting acceleration remain uncertain.
The researchers stressed that an intelligence explosion could accelerate the development of highly capable or superhuman AI systems and bring forward potential benefits, including medical cures and other transformative technologies but at the same time they pointed out three major risks.
Firstly the researchers said that an intelligence explosion could bring forward the risks of advanced AI and AI-enabled technologies, such as biological and cyber attacks, labor market disruption, and loss of control over AI systems themselves.
Secondly they mentioned that automating AI R&D could weaken human oversight. The paper said that humans could lose opportunities and expertise needed to identify and correct problems as they become less involved in the research.
The paper cited the OpenAI-Hugging Face incident as an example of potential loss of control risks.
“Without sufficient oversight, misaligned AI systems could “poison” the development of successors or bypass containment measures to act outside of their intended environments,” the paper said.
Thirdly the researchers said that an erosion of checks on power could weaken existing checks on power between states, companies and branches of government.
The researchers have further called on policymakers to take three broad steps. According to the researchers the policymakers should increase visibility into companies’ automation of AI R&D, they should develop ways to steer and constrain an intelligence explosion and they should be prepared to adapt to an intelligence explosion’s impacts.
Also Read: The Watchdog Problem: AI Agents Are Dulling the Humans Meant to Catch Their Mistakes
