The moment when artificial intelligence becomes smart enough to build even smarter AI on its own, a concept known as an “intelligence explosion,” was previously thought to be years away. However, a new update from the AI Futures Project suggests that this sci-fi scenario is now at our doorstep. According to the latest forecast, it is expected to happen as early as 2027.
On November 10, 2026, the six researchers behind the widely followed AI 2027 forecast published a document titled “Q3.5 Timelines Update: Looks Like the Intelligence Explosion Will Happen in 2027 After All.”
The authors now unanimously agree that a “Superhuman AI Researcher” will likely emerge between May and November 2027. They define this as an AI system so incredibly good at coding and scientific research that tech companies would rather rely entirely on automated AI than hire a team of the best human scientists.
Lead author and former OpenAI researcher Daniel Kokotajlo shared his concerns on X, noting that the intelligence explosion is happening five months ahead of their original AI 2027 scenario. He added that the rapid acceleration leaves him feeling scared, sad, and angry.
Why the timeline shifted so quickly
The sudden change in these predictions comes straight off the back of a major independent audit of the AI startup Anthropic. On November 5, 2026, Model Evaluation and Threat Research (METR) published a report after spending 11 weeks examining Anthropic’s internal model scores, training data, and coding activity.
METR researchers discovered that since October 2025, Anthropic has been improving its AI models at roughly twice the speed compared to its progress between March 2024 and October 2025.
What is even more surprising is that METR estimates about 80 per cent of this massive speed boost is happening because Anthropic engineers are heavily using their own unreleased AI models to write code, analyse data, and train newer models.
Taking these numbers into account, the AI Futures Project calculated that AI is already driving a 1.7x speed-up in overall AI research progress. In their older models, the authors had given less than a 4 per cent chance of seeing this level of acceleration by November 2026.
Anthropic confirms the rapid growth
The alarming data was confirmed by Anthropic itself on the very same day. In an official filing with the US Securities and Exchange Commission ahead of its planned IPO, the company stated that it had officially met its internal AI research and development acceleration threshold.
According to the company’s Responsible Scaling Policy, this specific threshold is crossed only when automated research allows the company to pack two years’ worth of normal AI progress into a single year.
A major threat to white-collar jobs
Alongside the arrival of a Superhuman AI Researcher, the report also brings bad news for human workers. The AI Futures Project updated its timeline for “Top-Expert-Dominating AI,” which refers to systems capable of doing virtually all white-collar human jobs.
Previously, researchers believed this level of full workplace automation would happen somewhere between late 2028 and early 2035. Now, they give a 35 to 65 per cent chance that such AI will arrive before the start of 2028.
Doubts and missing evidence
Despite the panic, both METR and the AI Futures Project admit that a lot of the hard evidence remains hidden from the public eye. To protect its trade secrets before going public, Anthropic redacted a large chunk of the internal data that METR used to reach its conclusions.
Furthermore, METR admitted there is substantial uncertainty in its 80 per cent estimate, clarifying that AI is mostly handling boring data tasks and not completely replacing human scientists just yet.
Several independent researchers on forums have also called the report an overreaction. Critics argue that giving human coders better AI tools naturally creates a one-time jump in productivity, which is very different from an AI being smart enough to invent new science all by itself.
Even with these doubts, the authors of the Q3.5 update are not taking any chances. They strongly warn that tech companies are currently not prepared to safely control a Superhuman AI Researcher, and are demanding an urgent, industry-wide pause on frontier AI training until an international safety agreement is signed.
– Ends
Leave a Reply