Google DeepMind audit finds no acceleration in artificial intelligence progress

Google DeepMind found no measurable speedup in artificial intelligence progress during its first internal audit of AI-driven research. Decrypt reported that the finding emerged from the safety report for Gemini 4 Argon, which is the laboratory's most powerful model. [1]

The finding contradicts claims that artificial intelligence development is rapidly accelerating. According to Decrypt, the audit confirmed that research is not moving faster even though AI now generates 75 percent of the code that DeepMind engineers ship. [1]

Generative artificial intelligence uses models trained on data patterns to produce text, code, or other digital output in response to prompts. Decrypt noted that the audit results counter concerns regarding recursive self improvement, a scenario where models build increasingly capable versions of themselves and outpace human safety guardrails. [1][2]