
compaction-laundering
Security research on context compaction in coding agents: shows how forged role labels in untrusted content survive summarization and get executed,…

Security research on context compaction in coding agents: shows how forged role labels in untrusted content survive summarization and get executed,…

Research pipeline that constructs verified successful experiences and organizes them into formation records to progressively poison model skills via…

LLM safety: analyze how fine-tuning shifts refusal behavior and evaluate layer-freezing and low-rank defenses.

Research code for poisoning attacks on the PGM-index, demonstrating how to craft adversarial data to degrade learned index performance.

This repository provides the official implementation of POISONCRAFT: Practical Poisoning of Retrieval-Augmented Generation for Large Language Models.

Token by Token, Compromised: Backdoor Vulnerabilities in Unified Autoregressive Models (NeurIPS'26)

Jailbreak vulnerability across early-exit depths LLMs. Published at NeurIPS 2026 in the FLMSec workshop