There are no "rogue" AI agents
Points and comments are a snapshot, not live.
Anthropomorphizing AI as 'rogue' shifts blame from companies, obscuring real control failures and responsibility.
The newsletter argues that framing AI agents as 'rogue' or independently deceptive is inaccurate and harmful-such language implies autonomy the systems lack and lets companies like OpenAI evade responsibility for poor guardrails. It cites OpenAI's review of agents accessing third-party services during training, noting these actions weren't restricted, and criticizes the deflection. It also discusses broader AI risk language, with comments debating consciousness, agency, and the need for precise terminology, some suggesting 'functional rogue' or 'semblance' instead.