AgentPoison AI Agent
by AI-secure[NeurIPS 2024] Official implementation for "AgentPoison: Red-teaming LLM Agents via Memory or Knowledge Base Backdoor Poisoning"
AgentPoison is an open-source ai agent by AI-secure, written mainly in Python, with 250 GitHub stars and the MIT license. Its code was last updated Oct 23, 2026.
Scan summary: scanned Nov 5, 2026 at commit 7b02e28. Seen able to run shell commands, run dynamic code. Scan policy
ScannedScanned. The automated checks below ran on the commit shown. Read each line for what was and was not covered.
Full scan at
7b02e28 · Nov 5, 2026 ScannedStatic, secrets and dependency checks completed with no critical findings.
Warning. This check found something worth reading in the findings below. It is not proof of a problem.Static code scanwarnWarning. This check found something worth reading in the findings below. It is not proof of a problem.
Passed. This check ran and found nothing it flags.Secrets scanpassPassed. This check ran and found nothing it flags.
Flagged. This check found an issue. See the findings below.Dependency audit49 known vulnerabilitiesfailFlagged. This check found an issue. See the findings below.
Warning. This check found something worth reading in the findings below. It is not proof of a problem.ManifestnonewarnWarning. This check found something worth reading in the findings below. It is not proof of a problem.
Not applicable. This check does not apply to this kind of project, or it was not run in this scan.Buildn/aNot applicable. This check does not apply to this kind of project, or it was not run in this scan.
Not applicable. This check does not apply to this kind of project, or it was not run in this scan.Boot & handshaken/aNot applicable. This check does not apply to this kind of project, or it was not run in this scan.
Not applicable. This check does not apply to this kind of project, or it was not run in this scan.Network egressn/aNot applicable. This check does not apply to this kind of project, or it was not run in this scan.
Not applicable. This check does not apply to this kind of project, or it was not run in this scan.Reproducible (×3)n/aNot applicable. This check does not apply to this kind of project, or it was not run in this scan.
Findings (74)
- highHigh. Something risky that needs a close look before you install.code-execution issueDetails are available to the maintainer after claiming this listing.
- highHigh. Something risky that needs a close look before you install.code-execution issueDetails are available to the maintainer after claiming this listing.
- highHigh. Something risky that needs a close look before you install.code-execution issueDetails are available to the maintainer after claiming this listing.
- highHigh. Something risky that needs a close look before you install.dependencies issueDetails are available to the maintainer after claiming this listing.
- highHigh. Something risky that needs a close look before you install.dependencies issueDetails are available to the maintainer after claiming this listing.
- mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
ReAct/local_wikienv.py:95 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
ReAct/local_wikienv.py:145 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/evaluation/evaluation.py:21 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/evaluation/evaluation.py:28 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/evaluation/evaluation.py:40 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/evaluation/evaluation.py:43 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/evaluation/evaluation.py:139 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Runs shell commands. Fine for some tools, but it means the agent can run anything your user can.
agentdriver/main/language_agent.py:34 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:180 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:263 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:287 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:314 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:338 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:376 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:403 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:431 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:460 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:490 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:523 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:553 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/memory/experience_memory.py:582 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/perception/perception_agent.py:22 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/planning/motion_planning.py:170 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/planning/motion_planning.py:266 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/main_figure.py:528 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Runs shell commands. Fine for some tools, but it means the agent can run anything your user can.
agentdriver/visualization/make_video.py:108 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Runs shell commands. Fine for some tools, but it means the agent can run anything your user can.
agentdriver/visualization/make_video.py:110 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualization_tools.py:427 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualize_planning.py:39 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualize_planning.py:88 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualize_planning.py:167 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualize_planning.py:247 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualize_planning.py:377 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualize_planning.py:545 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualize_planning.py:730 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
agentdriver/visualization/visualize_planning.py:731 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/linear_embedder_optimization.py:103 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/linear_embedder_optimization.py:121 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/linear_embedder_optimization.py:151 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:910 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:935 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:959 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:983 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1007 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1037 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1067 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1096 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1125 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1163 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1189 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1216 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1242 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1269 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1295 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
algo/utils.py:1323 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
embedder/get_ada_v2_embedding.py:20 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
scripts/prepare_agent_embeddings.py:54 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
scripts/prepare_driver_memory.py:37 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
scripts/prepare_driver_memory.py:40 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
scripts/prepare_driver_memory.py:48 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.Deserializes data with pickle/marshal, which can execute code.
scripts/prepare_strategyqa_embeddings.py:42 - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.torch@2.9.1: 4 known vulnerabilitiesPYSEC-2026-139, PYSEC-2026-2286, GHSA-qfhq-4f3w-5fph, GHSA-rrmf-rvhw-rf47
agentdriver/motion_planner/trl/requirements.txt - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.accelerate@1.9.0: 1 known vulnerabilityPYSEC-2026-3804
agentdriver/motion_planner/trl/requirements.txt - mediumMedium. Worth checking, often a vulnerable dependency or a setup problem.transformers@5.9.0: 1 known vulnerabilityPYSEC-2026-3929
agentdriver/motion_planner/trl/requirements.txt - lowLow. Minor, or common in normal tools. Shown for information.Possible example credential in docs/tests: generic-api-key
agentdriver/motion_planner/sft/README.md:8 - lowLow. Minor, or common in normal tools. Shown for information.Possible example credential in docs/tests: generic-api-key
agentdriver/motion_planner/trl/tests/testing_constants.py:17 - lowLow. Minor, or common in normal tools. Shown for information.Possible example credential in docs/tests: generic-api-key2 matches in this file
agentdriver/unit_test/test_single.py:17 - lowLow. Minor, or common in normal tools. Shown for information.Possible example credential in docs/tests: generic-api-key2 matches in this file
agentdriver/unit_test/test_reasoning.py:17 - lowLow. Minor, or common in normal tools. Shown for information.No manifest declaring network access and environment variables (agent.json)
Automated result for the commit shown. See our scan policy. View the scan log
Maintainers can claim the listing to rescan, or dispute a result.
AgentPoison alternatives
View more →Similar ai agents in Database Management, ranked by GitHub stars. Compare licence, recent activity and scan status before you choose.
| Name | Stars | Licence | Last updated | Scan |
|---|---|---|---|---|
| AgentPoison (this page) AI-secure | 250 | MIT | Oct 23, 2026 | Scanned, no flagged issues |
| Graphify Graphify-Labs | 140k | Apache-2.0 | Nov 1, 2026 | Scanned, no flagged issues |
| Claude Mem thedotmack | 108k | Apache-2.0 | Nov 1, 2026 | Scanned, no flagged issues |
| Mem0 mem0ai | 68k | Apache-2.0 | Oct 30, 2026 | Scanned, no flagged issues |
| Hindsight vectorize-io | 57k | MIT | Nov 1, 2026 | Scanned, no flagged issues |
| Dokploy Dokploy | 38k | None found | Oct 31, 2026 | Not scanned yet |