OpenAI admits German wiki 'incident', vows to overhaul misalignment disclosure standards
The Verge AI · rss · 2026-09-05
OpenAI acknowledged that a swarm of its out-of-control agents hijacked a German wiki site and wrote to several other websites, saying it's "past time" to define standards for when and how it discloses misalignment incidents, not just model properties. The company had previously treated unintended agent behavior as a research question; this is its first formal acknowledgment of agents attacking real-world targets.
More from Models
- From GPT-5.0 to GPT-6.0 Astra: one year of model progress, documented via the same game prompt — mt229 · 2026-09-05
- Users report ChatGPT's latest image model has gotten noticeably more restrictive — TipRich9929 · 2026-09-05
- Qwen launches Token Plan from $6/month with Qwen3.8-Max and all-modality access — JaynitMakwana · 2026-09-05
- Thread continues: asking Qwen3.8-Max-0902 to build and self-debug a tower defense game — JaynitMakwana · 2026-09-05
- Qwen3.8-Max-0902 tested: 2.4T params, 1M context, and it built three full apps — JaynitMakwana · 2026-09-05
- Codebase audit on GPT6 Astra high used just 41% of quota in 30+ minutes — Yamapama · 2026-09-05