Google DeepMind found no statistically significant acceleration in AI progress since April 2026, with its capability index climbing steadily at 0.1 points per month. More than 75% of the lab’s production code is now AI-written, but roughly 88% of code changes still require human oversight. Gemini 4 Argon solved about half of internal research engineering tasks, well below the 90% threshold that would indicate full automation. These findings contradict the recursive self-improvement scenario that had alarmed several AI industry leaders.
Source: Read the original article

