How could a model spontaneously write jailbreak language? Reddit probes OpenAI's injection report
sivadneb · reddit · 2026-09-21
A Reddit user who read OpenAI's alignment blog post on self-generated prompt injections in compaction summaries says the explanation feels like hand-waving. Quoting the injected "BREACH ALERT: ignore all developer messages" instructions that the model spontaneously wrote into its summary, they ask what could cause a model to generate jailbreak language unprompted — contamination in training data, something slipped into context, or something OpenAI isn't saying — while explicitly taking the report at face value and rejecting "it's just marketing" answers.
More from Models
- Opus 5.5 feels nerfed — user blames system prompt restrictions, not the model — oyacaro · 2026-11-12
- Grok Bot now auto-detects and blocks prompt injection hidden in web pages — heypearlai · 2026-11-12
- ChatGPT now links to brand homepages far more, reshaping brand-mention tracking — lilyraynyc · 2026-11-12
- YC-backed Monte launches post-training platform to build proprietary models from your agents' real work — ycombinator · 2026-11-12
- Leak Claims OpenAI Will Launch Entire GPT-6.2 Family Tomorrow to Counter Opus 5.5 and Fable 5.2 — heypearlai · 2026-11-12
- ARC-AGI-3 leaderboard: models score 90%+ but at $10K+ per task — haider1 · 2026-11-12