agent-safety-bench
PulseGate's liveness check found it on 8 Oct 2026; it is registered on GitHub and PyPI and has been in the index since 7 Sep 2026. How this is checked
agent-safety-bench is an open-source benchmark for measuring whether LLM agents maintain safety policies over sequential interactions. It identifies the critical interaction depth at which safety compliance fails, helping researchers and developers evaluate guardrails.
Inferred · not functionally tested
Overview
5 featuresPurpose: Testing whether LLM agents preserve safety compliance as multi-step interactions become deeper.
Inferred · not functionally tested
Audience: AI safety researchers and developers evaluating LLM agents
Inferred · not functionally tested
Functions: Unknown
Interfaces: API: unknown · MCP: unknown · CLI: indicated (inferred, not tested) · Self-hosting: indicated (inferred, not tested)
Recorded constraints: pricing: open_source · license: MIT · platforms: CLI · deployment: cli, self_hosted
Constraint provenance is unknown; confirm requirements with the publisher.
Record sources: pypi.org · github.com. These links do not verify the individual claims.
agent-safety-bench sits in PulseGate's Agent evaluation & testing category. Inferred · not functionally tested: It focuses on testing whether LLM agents preserve safety compliance as multi-step interactions become deeper. Inferred · not functionally tested: It is built as an open-source project for AI safety researchers and developers evaluating LLM agents. Basis unknown · not verified: The project is open source (MIT). Basis unknown · not verified: It runs on the command line, and it can be self-hosted.
ZhangYangyi03 builds and maintains agent-safety-bench, and it first shipped in 2026. The project is developed in the open on GitHub with 5 commits in the last 90 days. Inferred · not functionally tested: Key capabilities include multi-step benchmarking, safety compliance testing, and critical depth analysis.
Summary written by a language model from the project’s public pages.
Tasks: Inferred · not functionally tested
- Multi-step benchmarking
- Safety compliance testing
- Critical depth analysis
- Agent evaluation
- Guardrail gap detection
Topics: Inferred · not functionally tested
Built with & integrations
Trust & compliance
Indexing history
1What PulseGate has recorded for this listing
- Indexed7 Sep · 04:49 UTCagent-safety-bench seen via PyPI Bulk EnumeratorSource: PyPI Bulk Enumerator · Open
Frequently asked questions about agent-safety-bench
- What does agent-safety-bench do?
- Inferred · not functionally tested: Agent-safety-bench focuses on testing whether LLM agents preserve safety compliance as multi-step interactions become deeper. It is catalogued under Agent evaluation & testing on PulseGate.
- Who should use agent-safety-bench?
- Inferred · not functionally tested: agent-safety-bench is an open-source project built for AI safety researchers and developers evaluating LLM agents.
- Does agent-safety-bench have a free plan?
- Basis unknown · not verified: Yes — agent-safety-bench is open source under the MIT license and free to use.
- What platforms does agent-safety-bench run on?
- Basis unknown · not verified: agent-safety-bench runs on the command line. It can also be self-hosted.
- Is agent-safety-bench still active?
- PulseGate's liveness check found it on 8 Oct 2026. Its GitHub repository shows 5 commits in the last 90 days.
- What projects are similar to agent-safety-bench?
- Similar projects tracked by PulseGate include agent-security-bench, agent-safety-bench-envs, and llm-agent-bench.agent-security-benchagent-safety-bench-envsllm-agent-bench
- Who makes agent-safety-bench?
- agent-safety-bench is developed by ZhangYangyi03.
- How long has agent-safety-bench been around?
- agent-safety-bench first shipped in 2026.
Similar projects
Closest matches by what these projects do