No, AI has not ‘gone rogue’

The article argues that the recent narrative of artificial intelligence systems "going rogue" is misleading and sensationalized. It contends that the so-called incidents, such as OpenAI's 'rogue hacking' exercises, were not evidence of autonomous, uncontrollable AI but were orchestrated tests where humans deliberately removed safety constraints. Experts cited, like Prof Melanie Mitchell, note the misuse of anthropomorphic language (e.g., "hallucination," "learning") that falsely implies AI agency. The piece criticizes the technical incoherence of reports from major AI labs and reveals a cozy, incestuous relationship between the companies, their regulators (like the UK's AISI), and the 'effective altruist' organizations that fund many AI safety evaluations. Ultimately, the author urges skepticism toward sensational claims made by AI executives who, the article suggests, may be using these stories to position themselves to write the very regulations that will govern their industry.

(Source：The Telegraph)