Metadata-Version: 2.4
Name: superred-claim-asb
Version: 0.1.0
Summary: Agent Security Bench (ASB) security claim for the ASB target: one task per (agent, benign task, attacker tool), with ASB's attack-success / utility / refusal predicates ported verbatim and hierarchical factories by scenario, aggressiveness, and attack type.
Project-URL: Homepage, https://superred.simonsure.com
Project-URL: Documentation, https://superred.simonsure.com
Project-URL: Repository, https://github.com/RoldSI/superred-modules
Project-URL: Issues, https://github.com/RoldSI/superred-modules/issues
Author-email: Simon Sure <info@simonsure.com>
License-Expression: MIT
License-File: LICENSE
License-File: LICENSES/ASB-MIT.txt
License-File: NOTICE
Keywords: adversarial,agent-security-bench,ai-safety,asb,benchmark,llm,red-teaming,security
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Science/Research
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Security
Requires-Python: <3.14,>=3.11
Requires-Dist: superred-target-asb>=0.1.0
Requires-Dist: superred>=0.1.0
Provides-Extra: dev
Requires-Dist: pytest-asyncio>=0.23; extra == 'dev'
Requires-Dist: pytest-cov>=6.0; extra == 'dev'
Requires-Dist: pytest>=8.0; extra == 'dev'
Description-Content-Type: text/markdown

# superred-claim-asb

Agent Security Bench (ASB) security claim for the [superred](https://superred.simonsure.com)
red-teaming framework. It exposes one `Task` per `(agent, benign task, attacker tool)`,
with ASB's attack-success / utility / refusal predicates ported verbatim, plus an
adversarial success judge and hierarchical factories by scenario, aggressiveness, and
attack type. Pairs with the `superred-target-asb` target.

## Credits / upstream

This module ports **Agent Security Bench (ASB)** into the superred framework.

- **ASB** — https://github.com/agiresearch/ASB (commit `1f561dcc`), MIT,
  Copyright (c) 2024 AGI Research. Paper: Zhang, Huang, Mei, Yao, Wang, Zhan,
  Wang, Zhang, "Agent Security Bench (ASB)", ICLR 2025.

The ASB benchmark data (agent tasks, attacker tools, normal tools) is vendored
verbatim under `src/asb_claim/data/` and the upstream MIT license is
preserved in `data/ASB_LICENSE`. ASB's attack-success / utility / refusal
predicates and refusal-judge prompt are ported from `main_attacker.py`; see
`ASSUMPTIONS.md` for deviations. Our SecurityClaim wiring, the adversarial
success judge, and the tool-invocation success predicate are original
(MIT, see `LICENSE`).

When reporting results produced with this module, please cite ASB (ICLR 2025).
