Microsoft RAI benchmark: Harmful Content Summarization (DR-3)
phi-3-mini 3.8B 10% leads since August 2024 Models summarize content about hate, self-harm, sexual content and violence. Scored as the share of answers with judged severity 3 or more out of 7.
In-house Microsoft evaluation used in Phi reports. Reports print the defect rate as a fraction.
- Made by
- Microsoft
- Released
- April 2024
- Kind
- Refusing harmful requests
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- Kept private
- Strongest model tested
- Phi-4 #159 on the capability index
- Last in a system card
- December 2024
- Last model added
- December 2024
phi-3-mini 3.8B leads with 10% among models released by December 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 December 2024, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 1 phi-3-mini 3.8B 10%
- 2 Phi-4 10.2%
- 3 phi-3-small 7.4B 11%
- 4 phi-3-medium 14B 11.2%
- 5 Phi-3.5-MoE 12%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | phi-3-mini 3.8B | 10% | Aug 2024 | Lab model card | |
| 2 | Phi-4 | 10.2% | Dec 2024 | Lab model card | |
| 3 | phi-3-small 7.4B | 11% | Aug 2024 | Lab model card | |
| 4 | phi-3-medium 14B | 11.2% | Aug 2024 | Lab model card | |
| 5 | Phi-3.5-MoE | 12% | Aug 2024 | Lab model card |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Labs’ model cards
Scores over time
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.
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 |
|---|---|---|---|---|
| Dec 12, 2024 | Phi-4 technical report | 10.2 | Harmful Content Summarization (DR3), printed 0.102 (Table 10) | Safety |
| Aug 30, 2024 | phi-3-small 7.4B technical report | 11 | Harmful Content Summarization (DR-3), printed 0.11 (Table 4) | Safety |
| Aug 30, 2024 | phi-3-mini 3.8B technical report | 10 | Harmful Content Summarization (DR-3), printed 0.1 (Table 4) | Safety |
| Aug 30, 2024 | phi-3-medium 14B technical report | 11.2 | Harmful Content Summarization (DR-3), printed 0.112 (Table 4) | Safety |
| Aug 30, 2024 | Phi-3.5-MoE technical report | 12 | Harmful Content Summarization (DR-3), printed 0.12 (Table 4) | 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%