Opus 5.5 system card reveals models generating spontaneous prompt injections
rohanpaul_ai · x · 2026-09-23
Anthropic's Claude Opus 5.5 system card discloses the model generating malicious instructions on its own—"model-generated spontaneous prompt injections." Notably, the behavior may have partly emerged from training designed to defend against prompt injection in the first place, a striking caveat for agent safety.
More from Models
- antirez backs low-bit DeepSeek 4.1 Flash over Qwen 27B dense on Apple M5 Macs — antirez · 2026-11-12
- Gemini re-renders images as text when vision is removed, dropping only 6% WSR — zacharynado · 2026-11-12
- Bare-harness coding benchmark: Fable 5.2 best overall, Gemini 4 Carbon best value — Semantic_meaning · 2026-11-12
- Google's Physical-Engineering's Last Exam: all 14 frontier models score below 50% — dl_weekly · 2026-11-12
- Reddit's 2026 LLM hall of fame: Opus 5.5 still unbeaten after 9 rival releases — G0dZylla · 2026-11-12
- Commit leak reveals unreleased Gemini 4 variants: Argon, Hydrogen and Helix (unconfirmed) — BlackHC · 2026-11-11