
ShadowMem
Defensive framework that maintains a safety-focused shadow memory to detect and block prompt-injection and long-horizon threats against LLM agents…

Defensive framework that maintains a safety-focused shadow memory to detect and block prompt-injection and long-horizon threats against LLM agents…
Measuring Open Privilege in Agent Defenses

Experiments for control-token chain-of-thought suppression and parser-leniency attacks on tool-using LLM agents

Proof-of-concept that poisons MLflow registered models via the REST API, embedding a malicious pickle to trigger RCE when the model is loaded.

Proof-of-concept and research repository for CVE-2024-37054, an unsafe deserialization flaw in MLflow PyFunc model loading that can lead to remote…

Adaptive two-stage Layer 4 DDoS mitigation gateway using behavioral traffic analysis, Random Forest classification, and kernel-level ipset/iptables…

Bidirectional token-classification model for PII detection and masking in text, with CLI for redaction, evaluation, and finetuning on-premises.

Research code for a gray-box trojan attack that flips a single KV-cache bit in fine-tuned LLM classifiers and measures per-class attack success rate.

Pre-execution action-auditing defense that detects and masks indirect prompt injection in tool-using LLM agents using embedding retrieval and…

Multi-agent automated context management for long horizon tasks in local AI Agents

This repository contains the official implementation of the paper "[Safety in Batches? Understanding and Mitigating Safety Failures in Batch…

Robust audio watermarking framework embedding binary messages into magnitude spectrograms, with differentiable attack simulation, Q-Former pooling,…

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

Research code for extracting and training safety-awareness directions in multimodal LLMs to improve refusal behavior while limiting benign-task drift.

ML-driven threat detection and continuous monitoring platform built for federal zero trust architectures.

A Python pickling decompiler and static analyzer

Linux system-call monitor using ptrace to trace file, process, network, and memory activity, with namespace isolation and machine learning…

Python library for adversarial machine learning security, enabling red and blue teams to run evasion, poisoning, extraction, and inference attacks…