7 min read
Research Three papers published in 2026 confirm what practitioners suspected: LLM safety alignment is structurally shallow, and fine-tuning APIs are the widest open bypass.
Three papers published in 2026 confirm what practitioners suspected: LLM safety alignment is structurally shallow, and fine-tuning APIs are the widest open bypass.
Backdoor behaviours introduced into LLMs during fine-tuning can persist through subsequent safety alignment including RLHF and adversarial training, surviving standard red-teaming.
Model inversion and training data extraction attacks allow adversaries to recover PII, proprietary data, and trade secrets from fine-tuned LLMs exposed via API.