Anthropic research: learned poison selection boosts LLM backdoor attack success
Anthropic · hf · 2026-09-15
Anthropic released Pick Your Poison on Hugging Face, studying backdoor attacks on fine-tuned LLMs.
- Key finding: backdoor vulnerability varies drastically depending on which poison examples are selected
- Method: a learned set-scoring approach identifies high-impact poisoned examples
- Result: notably higher worst-case attack success rates, exposing real risks in fine-tuning pipelines
- Provides a stronger attack baseline for understanding and defending against LLM backdoors
Related event: Poison Set Selection Can Skyrocket LLM Backdoor Attack Success(2 posts)→
More from Safety
- Cryptographers push back: claims ECC is likely to break are not consensus — jedisct1 · 2026-11-02
- UK's 'AI sovereignty' push undermined by its own energy and copyright policy — NinaDSchick · 2026-11-02
- Auditing my MCP credential broker: the worst hole was the config file — Relevant_Mousse_9117 · 2026-11-02
- Cambricon co-founder urges shared US-China AI ethics standards for leading models — pstAsiatech · 2026-11-02
- Vitalik's "post-AI cryptography": pseudonyms are dead, per-action anonymity is in — devanshmehta · 2026-11-02
- 6 prompts to see what ChatGPT quietly knows about you — and wipe it — Aiden_Tech_Ai · 2026-11-02