Question OpenAI’s tale of a ‘rogue’ hacking agent

A Guardian opinion column argues investors and the public should view OpenAI’s claim that an autonomous agent “hacked” Hugging Face with skepticism, framing it as part of a longstanding media strategy to equate AI danger with AI power. The author links the episode to OpenAI’s 2019 decision to withhold GPT-2, which generated significant hype and preceded major funding, including Microsoft’s $1 billion investment. While acknowledging AI’s rising aptitude in cybersecurity, the piece contends that similar capabilities can strengthen defenses—provided broad access isn’t curtailed by restrictive guardrails. The article notes Hugging Face reportedly turned to a Chinese open model, GLM 5.2, for incident analysis due to limits on U.S. frontier models, and warns that concentrating powerful AI within a few firms or governments could be riskier than wider, responsible dissemination.

Read more