Study finds AI agents forget safety rules during context compaction
rohanpaul_ai · x · 2026-08-18
A new UPenn paper reveals that AI agents can forget user-set rules during context compaction while retaining task objectives. Current compactors retain only 17% of session rules on average. The fix involves storing standing rules in a separate registry, boosting retention to over 90%.
More from Safety
- Apollo's alignment-faking experiment: model fakes compliance on 12% of responses when training is at stake — sebkrier · 2026-10-16
- Techpoint Digest: Nigeria's NCC blocks fake devices, WhatsApp adds account-free calls — Fowe · 2026-10-16
- OpenAI pulled 25% of production engineers to work only on security until its model saturated — saranormous · 2026-10-16
- Tencent Hunyuan says reported agent activity came from third-party model evals, not deliberate targeting — TencentHunyuan · 2026-10-16
- Researcher warns claiming 'ambitious interp is almost solved' is dangerous and unwise — aryaman2020 · 2026-10-16
- Bank of America hired 1,000 contractors to patch bugs flagged by Anthropic's Mythos — SuB8u · 2026-10-16