foundation models

May 22, 2026 5 min read

Sleeper Agents in Fine-Tuned LLMs

Backdoor behaviours introduced into LLMs during fine-tuning can persist through subsequent safety alignment including RLHF and adversarial training, surviving standard red-teaming.

img of Membership Inference Attacks on LLMs

May 19, 2026 5 min read

Research

Membership Inference Attacks on LLMs

Training data can be reconstructed from foundation model weights with significantly higher accuracy than previously reported, with implications for GDPR compliance and IP protection.