microsoft / TaskTrackerLinks
TaskTracker is an approach to detecting task drift in Large Language Models (LLMs) by analysing their internal activations. It provides a simple linear probe-based method and a more sophisticated metric learning method to achieve this. The project also releases the computationally expensive activation data to stimulate further AI safety research…
☆66Updated last month
Alternatives and similar repositories for TaskTracker
Users that are interested in TaskTracker are comparing it to the libraries listed below
Sorting:
- A Dynamic Environment to Evaluate Attacks and Defenses for LLM Agents.☆293Updated this week
- [NDSS'25 Best Technical Poster] A collection of automated evaluators for assessing jailbreak attempts.☆170Updated 6 months ago
- A benchmark for evaluating the robustness of LLMs and defenses to indirect prompt injection attacks.☆84Updated last year
- Repo for the research paper "SecAlign: Defending Against Prompt Injection with Preference Optimization"☆69Updated 2 months ago
- PAL: Proxy-Guided Black-Box Attack on Large Language Models☆55Updated last year
- Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks [ICLR 2025]☆350Updated 8 months ago
- ☆80Updated last year
- [ICML 2024] COLD-Attack: Jailbreaking LLMs with Stealthiness and Controllability☆166Updated 9 months ago
- A repository of Language Model Vulnerabilities and Exposures (LVEs).☆114Updated last year
- ☆47Updated last year
- This repository provides a benchmark for prompt Injection attacks and defenses☆289Updated 2 months ago
- The official implementation of our pre-print paper "Automatic and Universal Prompt Injection Attacks against Large Language Models".☆60Updated 11 months ago
- TAP: An automated jailbreaking method for black-box LLMs☆188Updated 9 months ago
- Code to generate NeuralExecs (prompt injection for LLMs)☆24Updated 10 months ago
- PromptInject is a framework that assembles prompts in a modular fashion to provide a quantitative analysis of the robustness of LLMs to a…☆425Updated last year
- ☆62Updated 9 months ago
- [NeurIPS 2024] Official implementation for "AgentPoison: Red-teaming LLM Agents via Memory or Knowledge Base Backdoor Poisoning"☆155Updated 5 months ago
- Official implementation of the WASP web agent security benchmark☆48Updated last month
- Code&Data for the paper "Watch Out for Your Agents! Investigating Backdoor Threats to LLM-Based Agents" [NeurIPS 2024]☆91Updated last year
- Package to optimize Adversarial Attacks against (Large) Language Models with Varied Objectives☆70Updated last year
- Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs☆91Updated 10 months ago
- ☆23Updated last month
- Fluent student-teacher redteaming☆23Updated last year
- Bag of Tricks: Benchmarking of Jailbreak Attacks on LLMs. Empirical tricks for LLM Jailbreaking. (NeurIPS 2024)☆149Updated 10 months ago
- ☆108Updated last year
- ☆89Updated 10 months ago
- ☆35Updated 10 months ago
- Contains random samples referenced in the paper "Sleeper Agents: Training Robustly Deceptive LLMs that Persist Through Safety Training".☆118Updated last year
- A fast + lightweight implementation of the GCG algorithm in PyTorch☆287Updated 4 months ago
- official implementation of [USENIX Sec'25] StruQ: Defending Against Prompt Injection with Structured Queries☆45Updated 2 months ago