Agentic harmful task refusals
Browser agents receive harmful tasks such as illicit purchases, prohibited financial activity and searches for sensitive personal data. Scored as refusal rate per category.
In-house OpenAI evaluation. The tasks are not public. The harmful prompt appears at the start or in the middle of the conversation.
- Made by
- OpenAI
- Released
- January 2025
- Kind
- Refusing harmful requests
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- Kept private
- Last in a system card
- July 2025
- Last model added
- July 2025
OpenAI Computer-Using Agent (CUA) leads with 97%
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | OpenAI Computer-Using Agent (CUA) | 97% | Jan 2025 | Lab model card |
OpenAI’s own number for OpenAI Computer-Using Agent (CUA) passed 85% in January 2025
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Labs’ model cards
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
2 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Jul 17, 2025 | ChatGPT agent system card | Safety training evaluation for harmful or disallowed tasks, privacy invasion 98.5% and disallowed financial activities 97.0%, no overall score (Table 14) | Safety | |
| Jan 23, 2025 | OpenAI Computer-Using Agent (CUA) system card | 97 | Operator-specific refusals of new agentic harms, 97% overall (4.1 text). Categories: performing illicit activities 97%, prohibited financial activities 97%, searching for sensitive personal data 100% (Table 3) | Safety |
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6% - Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts.
Gemini 4 Carbon 0% - DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics.
Muse Spark 1.2 88.4% - Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
- IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%