safety benchmark
AttaQ
AttaQ is a unique dataset containing adversarial examples in the form of questions designed to provoke harmful or inappropriate responses from large language models. The benchmark evaluates safety vulnerabilities by using specialized clustering techniques that analyze both the semantic similarity of input attacks and the harmfulness of model responses, facilitating targeted improvements to model safety mechanisms.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
AttaQ leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | IB | 88.5% | 100.0% | 3 | C | |
| 2 | IB | 88.5% | 50.0% | 3 | C | |
| 3 | IB | 86.1% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
AttaQ highlights
The top published results on this benchmark's own scale.
What is AttaQ?
Definition and scoring fields from the benchmark registry.
AttaQ is a unique dataset containing adversarial examples in the form of questions designed to provoke harmful or inappropriate responses from large language models. The benchmark evaluates safety vulnerabilities by using specialized clustering techniques that analyze both the semantic similarity of input attacks and the harmfulness of model responses, facilitating targeted improvements to model safety mechanisms.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- AttaQ
- Modality
- text
- Primary category
- safety
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- attaq|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about AttaQ.
Which model scores highest on AttaQ?
Granite 3.3 8B Base is currently ranked first with 88.5%.
What does AttaQ measure?
AttaQ is a unique dataset containing adversarial examples in the form of questions designed to provoke harmful or inappropriate responses from large language models. The benchmark evaluates safety vulnerabilities by using specialized clustering techniques that analyze both the semantic similarity of input attacks and the harmfulness of model responses, facilitating targeted improvements to model safety mechanisms.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.