Google DeepMind said in the security report for Gemini 4 Argon that its first broad empirical assessment of machine learning R&D automation found no significant end-to-end acceleration in AI research driven by AI itself. The finding stands even though AI now writes most of the production code shipped by engineers at the lab.
Haydn Belfield, a researcher involved in the assessment, said on X that there is currently no meaningful end-to-end ML R&D acceleration, which in his view keeps the present risk from AI-driven ML R&D relatively low. To track capability progress, the team built an internal index called GCI v1, which combines scores from 123 AI models across 92 benchmarks into a single value.
The report said frontier Gemini capability has been rising at a steady pace of about 0.1 GCI points per month. Gemini 1.5 Pro was close to 0.5 by late 2024, Gemini 3.8 Flash was slightly above 2.9 in September 2026, and Argon reached about 3.3, above the trend line. Belfield attributed that gap to increased compute and organizational investment rather than AI-led acceleration. The report also said more than 75% of DeepMind production code is written by AI, while roughly 88% of code merge requests still require repeated human guidance or direct human writing.
Google DeepMind said in the Gemini 4 Argon security report that its first overall empirical assessment of machine learning R&D automation found that AI has not accelerated AI research itself, even though AI now writes most of the code delivered by engineers at the lab.
Haydn Belfield, a researcher involved in the assessment, wrote on X that there is currently no significant end-to-end ML R&D acceleration, which means the risk from AI-driven ML R&D remains low for now.
GCI v1 tracks capability gains across models and benchmarks
The team built an internal index, GCI v1, by combining scores from 123 AI models across 92 benchmarks into a single number. According to the report, frontier Gemini capability has been improving at a stable rate of about 0.1 GCI points per month.
Gemini 1.5 Pro was close to 0.5 by late 2024. Gemini 3.8 Flash was slightly above 2.9 in September 2026. Argon came in at about 3.3, above the trend line. Belfield said that outperformance reflected higher compute and greater organizational investment, not acceleration driven by AI.
AI writes most production code, but humans still handle much of the process
The report said more than 75% of DeepMind production code is written by AI. The share of fully delegated code merge requests rose from 7.5% to 11.9% between April and September. Even so, about 88% of requests still required multiple rounds of human guidance or direct human-written code.
Automation performs best on well-defined execution work
A survey of 12 senior researchers found that automation performs best on clear and verifiable execution tasks. It performs worst on open-ended scientific ideation, new experiment design, and diagnosing unstable training runs.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
