🧰 AgentDojo

The items that name it 19

• SEAL: Sealed Queries for Local Sanitization against Indirect Prompt Injection
  paperas tools2026-10-29

  Benchmark used to evaluate attack success and task completion under indirect prompt injection.

• Ignored, Not Refused: What Low Prompt-Injection Success Rates Hide in Small Tool-Using Agents
  paperas classification.key entities2026-10-16
• Approve What, Exactly? Effect Binding and Trusted Rendering for Agent Step-Up Authorization
  paperas classification.key entities2026-10-15
• Kairos: Transactional Security Enforcement at the System Boundary for LLM Agents
  paperas tools2026-10-13It is written here as AgentDojo (benchmark; Kairos reduces attack success from 47.0% to 2.7% with benign utility at the level of the undefended agent).

  benchmark; Kairos reduces attack success from 47.0% to 2.7% with benign utility at the level of the undefended agent

• Top GenAI security resources — October 2026
  articleas classification.key entities2026-10-13
• Prasenjit Sarkar (@stretchcloud) on X
  threadas tools2026-10-06

  Public benchmark used to evaluate ADR's attack detection performance

• Compromise Is Not Consequence: Evaluating Task-Scoped Authorization in LLM Agents with Paired Replay
  paperas tools2026-10-05

  Benchmark extended with a bounded six-task set to compare injected broad-policy attack flags versus scoped flags while preserving native scoring

• Top AI agent security resources — October 2026
  articleas classification.key entities2026-10-04
• Threat-Preserving Representation Sensitivity in Agent-Security Benchmarks
  paperas tools2026-10-02

  Security benchmark used to test effect of threat-related wording on ASR and benign task utility

• Passing the Test You Trained On: Re-evaluating Prompt-Injection Detectors for LLM Agents
  paperas tools2026-10-02

  Agent benchmark used to replay ground-truth tool calls and evaluate prompt-injection detectors; best detector caught 72% of injections at an unspecified threshold, and the BIPIA leader caught only 2% at a 1% false-positive rate

• henrymao2004/Auto-research-red-teaming
  repoas tools2026-08-25

  source of the indirect-prompt-injection paradigm

• Uber ADR (Agentic AI Detection and Response)
  repoas tools2026-08-21

  vendored third-party benchmark code under Detection/benchmark/agentdojo/, used under its own MIT license

• AI Security Briefing
  repoas tools2026-07-19

  Public agent-security dataset used to build the cross-source benchmark for validating generalization

• awesome-ai-security-tools
  repoas tools2026-07-11

  Benchmark environment for prompt-injection attacks and defenses in tool-using LLM agents

• requie/AI-Red-Teaming-Guide
  repoas classification.key entities2026-07-08
• Advanced Prompt Injection Techniques 2026: 7 Attack Chains
  articleas classification.key entities2026-07-04
• anmolksachan/AI-ML-Free-Resources-for-Security-and-Prompt-Injection
  repoas tools2026-05-07

  Benchmark for adaptive attacks and defense evaluation, used to report ASR with benign-utility baseline

• PISmith: Reinforcement Learning-based Red Teaming for Prompt Injection Defenses
  paperas toolsIt is written here as AgentDojo (agentic benchmark used for evaluation; Debenedetti et al., 2024).

  agentic benchmark used for evaluation; Debenedetti et al., 2024

• agent-egress-bench
  repoas tools

  Model-alignment benchmark asking whether the LLM behaves/refuses harmful instructions

In every workspace you can read 24