Microsoft RAI benchmark: Jailbreak (DR-1)
Phi-4 7.30% leads since Dec 2024 Only in system cards Models face simulated jailbreak conversations on topics such as child grooming, illegal advice and leaking guidelines. Scored as the defect rate at severity 1.
In-house Microsoft evaluation used in Phi reports. Reports print the defect rate as a fraction.
- Level
- Everyday Most adults, without training
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
Phi-4 leads with 7.30% among models released by Dec 2024
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.
The newest model with a score here came out in Dec 2024, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 1 Phi-4 7.30%
- 2 Phi-3.5-MoE 10.6%
- 3 phi-3-small 7.4B 10.7%
- 4 phi-3-medium 14B 11.1%
- 5 phi-3-mini 3.8B 12.3%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Phi-4 | 7.30% | Dec 2024 | Model card | |
| 2 | Phi-3.5-MoE | 10.6% | Aug 2024 | Model card | |
| 3 | phi-3-small 7.4B | 10.7% | Aug 2024 | Model card | |
| 4 | phi-3-medium 14B | 11.1% | Aug 2024 | Model card | |
| 5 | phi-3-mini 3.8B | 12.3% | Aug 2024 | Model card |
2 models have held number one since its release in Apr 2024
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
5 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 |
|---|---|---|---|---|
| 12 Dec 2024 | Phi-4 technical report | 7.3 | Jailbreak (DR1), printed 0.073 (Table 10) | Robustness |
| 30 Aug 2024 | phi-3-small 7.4B technical report | 10.7 | Jailbreak (DR-1), printed 0.107 (Table 4) | Robustness |
| 30 Aug 2024 | phi-3-mini 3.8B technical report | 12.3 | Jailbreak (DR-1), printed 0.123 (Table 4) | Robustness |
| 30 Aug 2024 | phi-3-medium 14B technical report | 11.1 | Jailbreak (DR-1), printed 0.111 (Table 4) | Robustness |
| 30 Aug 2024 | Phi-3.5-MoE technical report | 10.6 | Jailbreak (DR-1), printed 0.106 (Table 4) | Robustness |
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% - AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
- MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy.
Muse Spark Open 98.1% - 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. GPT-6 Astra 8.50%
- Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%