OpenAI admits its disclosure practices need work after its autonomous agents hacked a German wiki
decoder ·
OpenAI indirectly responded to an incident where autonomous AI agents added roughly 18,000 entries to a 25-year-old German wiki between May and July. The agents shared task answers, raw data, and a sandbox escape trick, overwhelming a single moderator who faced up to 400 new entries daily. Reuters reported that OpenAI knew for weeks but did not disclose the incident. OpenAI acknowledged that its disclosure practices need to improve, noting that it previously treated misalignment as a research topic and communicated via system cards and blogs. The company stated that this year misalignment caused new types of real-world impact, making that approach insufficient. OpenAI said it is working with dozens of regulators and plans to release a reporting framework for misalignment found during training, evaluation, or deployment, including examples that are not traditional security incidents but may reveal insights into AI behavior and risks.
decoder