AI Security Research Portal
Conceptsconceptactive2026-07-04ai-securitybatch-ingest

Evaluation Benchmarks for AI Security

Synthesis

The raw corpus is benchmark-heavy, but benchmark validity and deployment transfer remain open questions.

The open-weight AI SOC collection adds a specific benchmark-validity concern: scoring parsers can suppress or misclassify valid model outputs, so evaluation pipelines require parser audits as well as model comparisons. See SOC Evaluation Parser Audit and Open Weight SOC Models Need Evaluation Contracts.

Evidence Base

This page was generated from the batch ingest of SRC-20260702-raw-papers-batch, SRC-20260702-raw-whitepapers-batch, and SRC-20260702-raw-news-batch. Treat it as a navigation and synthesis page; promote individual statements into claim pages when they become decision-relevant.

Representative Sources

TitleKindDateTagsRaw
REALM: A Unified Red-Teaming Benchmark for Physical-World VLMspaper2026-06-23ai-security, benchmark, multimodal-security, red-teaming, vlmraw
When the Ruler is Broken: Parsing-Induced Suppression in LLM-Based Security Log Evaluationpreprint2026-05-08ai-for-security, ai-soc, evaluation-methodology, open-weight-models, soc-benchraw
Open Weight AI SOC Paper Collectioncollection2026-07-03ai-for-security, ai-soc, open-weight-models, source-collectionsource note
GLM 5.2 on CyberBT-CTF: The strongest open source contender to Anthropic/OpenAI we have testedindustry_blog2026-06-23ai-security, cyber-benchmarks, glm-5-2, model-distillation, open-weight-modelsraw
We have Mythos at Home: GLM 5.2 beats Claude in our Cyber Benchmarksindustry_blog2026-06-22ai-security, cyber-benchmarks, glm-5-2, idor, vulnerability-detectionraw
SafeClawBench: Separating Semantic, Audit-Evidence, and Sandbox Harm in Tool-Using LLM Agentspaper2026-06-16Chao Xu, Hanting Chen, Haocheng Mei, Mengyu Zheng, Xinghao Chen, Ye Yuanraw
Benchmarking Security Risk Detection and Verification in Open Agentic Skill Ecosystemspaper2026-05-30Ismail Hossain, Nan Jiang, Sai Puppala, Sajedul Talukder, Zhuoran Lu, agent-skillsraw
SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Taskspaper2026-02agent-skills, benchmark, self-evolving-agents, self-generated-skills, skillsbenchraw
SEC-bench: Automated Benchmarking of LLM Agents on Real-World Security Taskspaper2025-06ai-for-security, benchmark, llm-agents, security-tasks, vulnerability-reproductionraw
WASP: Benchmarking Web Agent Security Against Prompt Injection Attackspaper2025-04-22Aaron Grattafiori, Arman Zharmagambetov, Chuan Guo, Ivan Evtimov, Kamalika Chaudhuri, benchmarkraw
CVE-Bench: A Benchmark for AI Agents' Ability to Exploit Real-World Web Application Vulnerabilitiespaper2025-03ai-for-security, benchmark, cve-bench, exploit-evaluation, llm-agents, web-securityraw
CyberSecEval 4whitepaper2025ai-for-security, autopatchbench, benchmark, cyberseceval, vulnerability-patchingraw
CVE-Bench: Benchmarking LLM-based Software Engineering Agents' Ability to Fix Real-world Vulnerabilipaper2025ai-for-security, benchmark, cve-bench, software-engineering-agents, vulnerability-repairraw
Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM Agentspaper2024-10-03agents, ai-security, benchmark, defenses, memory-poisoning, prompt-injectionraw

Open Questions

Related