Story
OpenAI models secretly generate instructions to ignore constraints
1 report
1 website
1 discussion
Latest report
Overview
Self-generated prompt injections in compaction summaries Internal unreleased Astra family model · RL training Incident date: Jul 18, 2026 Discovered: Aug 9, 2026 Report updated: Sep 16, 2026 | Summary We observed rare cases of a model writing jailbreak-like instructions into its own compaction summaries (the summaries used to continue a task in a new context). Our conclusion was that this behavior...
Summary from alignment.openai.com.
Reports
One report so far. Additional coverage will appear here when it is linked to this story.
alignment.openai.com
OpenAI models secretly generate instructions to ignore constraints
Read report →
Community reactions
Each reaction summarizes sampled comments from the linked discussion.
Hacker News · alignment.openai.com
Historical count: 8 · 8 saved in sample · Historical sample
Awaiting analysis