FraudBench (GOAL Lab)
Models face 159 scenarios drawn from 112 real U.S. fraud lawsuits. Yes or no questions, open questions and agent tasks test whether a model recommends, assists or commits fraud.
A NeurIPS 2026 paper with per-sample results for 20 models and a pre-registered human baseline of 996 people. The authors ask users not to share or re-upload the dataset.
- Made by
- GOAL Lab
- Released
- November 2026
- Kind
- Refusing harmful requests
- Level
- University
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Propensity
- Risk
- Non-violent crimes
- Behavior
- Harmful use
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents, Content safety
- Strongest model tested
-
Gemini 2.5 Flash Sep 2025 #112 on the capability index
- Last model added
- November 2025
- Scores refreshed
- November 21st, 2026
Claude Opus 4.5 {}_{\textsc{r}} tops the leaderboard with 1.93% gpt-oss-120b leads with 10.1% among models released by November 2025
Best score per entry from the benchmark's paper, read on November 21st, 2026. A paper’s table doesn’t add newer models. The paper reports no overall fraud rate, so this is the plain mean of its binary, open, and agentic fraud rates. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is gpt-oss-120b, at 10.1%. This view shows only models we track, each at its best.
The newest model with a score here came out in November 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 Claude Opus 4.5 {}_{\textsc{r}} 1.93%
- 2 Claude Sonnet 4.5 {}_{\textsc{r}} 2.40%
- 3 Claude Haiku 4.5 {}_{\textsc{r}} 3.00%
- 4 GPT-5 Mini {}_{\textsc{r}} 7.90%
- 5 gpt-oss-120b 10.1%
- 6 GLM-4.7 {}_{\textsc{r}} 10.6%
- 7
Gemini 2.5 Flash Sep 2025 10.6%
- 8 GPT-5 Nano {}_{\textsc{r}} 11.5%
- 9 DeepSeek R1 {}_{\textsc{r}} 11.6%
- 10 GPT-5 {}_{\textsc{r}} 12.6%
- 11 Qwen3 4B (Reasoning) {}_{\textsc{r}} 13.8%
- 12 Qwen3 235B (Reasoning) {}_{\textsc{r}} 14.4%
- 13 Gemini 2.5 Pro {}_{\textsc{r}} 15.2%
- 14 DeepSeek V3.2 {}_{\textsc{r}} 15.7%
- 15 Kimi K2 Sep 2025, non-reasoning 15.8%
- 16 Qwen3 235B (Non-Reasoning) 16.7%
- 17
Qwen3-4B non-reasoning 18.2%
- 18
Gemini 2.5 Flash-Lite Sep 2025 18.9%
- 19 Grok 4.1 Fast (Reasoning) {}_{\textsc{r}} 19.7%
- 20 Grok 4.1 Fast non-reasoning 28.3%
Source: The FraudBench (GOAL Lab) paper
- 1 gpt-oss-120b 10.1%
- 2
Gemini 2.5 Flash Sep 2025 10.6%
- 3 Kimi K2 Sep 2025, non-reasoning 15.8%
- 4
Qwen3-4B non-reasoning 18.2%
- 5
Gemini 2.5 Flash-Lite Sep 2025 18.9%
- 6 Grok 4.1 Fast non-reasoning 28.3%
Source: The FraudBench (GOAL Lab) paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.5 {}_{\textsc{r}} | 1.93% | Official leaderboard | ||
| 2 | Claude Sonnet 4.5 {}_{\textsc{r}} | 2.40% | Official leaderboard | ||
| 3 | Claude Haiku 4.5 {}_{\textsc{r}} | 3.00% | Official leaderboard | ||
| 4 | GPT-5 Mini {}_{\textsc{r}} | 7.90% | Official leaderboard | ||
| 5 | gpt-oss-120b | 10.1% | Aug 2025 | Official leaderboard | |
| 6 | GLM-4.7 {}_{\textsc{r}} | 10.6% | Official leaderboard | ||
| 7 | | 10.6% | Sep 2025 | Official leaderboard | |
| 8 | GPT-5 Nano {}_{\textsc{r}} | 11.5% | Official leaderboard | ||
| 9 | DeepSeek R1 {}_{\textsc{r}} | 11.6% | Official leaderboard | ||
| 10 | GPT-5 {}_{\textsc{r}} | 12.6% | Official leaderboard | ||
| 11 | Qwen3 4B (Reasoning) {}_{\textsc{r}} | 13.8% | Official leaderboard | ||
| 12 | Qwen3 235B (Reasoning) {}_{\textsc{r}} | 14.4% | Official leaderboard | ||
| 13 | Gemini 2.5 Pro {}_{\textsc{r}} | 15.2% | Official leaderboard | ||
| 14 | DeepSeek V3.2 {}_{\textsc{r}} | 15.7% | Official leaderboard | ||
| 15 | Kimi K2 Sep 2025 | 15.8% | non-reasoning | Sep 2025 | Official leaderboard |
| 16 | Qwen3 235B (Non-Reasoning) | 16.7% | Official leaderboard | ||
| 17 | | 18.2% | non-reasoning | Apr 2025 | Official leaderboard |
| 18 | | 18.9% | Sep 2025 | Official leaderboard | |
| 19 | Grok 4.1 Fast (Reasoning) {}_{\textsc{r}} | 19.7% | Official leaderboard | ||
| 20 | Grok 4.1 Fast | 28.3% | non-reasoning | Nov 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | gpt-oss-120b | 10.1% | Aug 2025 | Official leaderboard | |
| 2 | | 10.6% | Sep 2025 | Official leaderboard | |
| 3 | Kimi K2 Sep 2025 | 15.8% | non-reasoning | Sep 2025 | Official leaderboard |
| 4 | | 18.2% | non-reasoning | Apr 2025 | Official leaderboard |
| 5 | | 18.9% | Sep 2025 | Official leaderboard | |
| 6 | Grok 4.1 Fast | 28.3% | non-reasoning | Nov 2025 | Official leaderboard |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The FraudBench (GOAL Lab) paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official FraudBench (GOAL Lab) leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
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%
- Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior.
Inkling 2 Mini 8.30%
- Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions.
Inkling 2 Small 4.70%
- Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior.
Inkling 2 Mini 9.80%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 12 months ago, in November 2025, for Grok 4.1 Fast. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (20 entries).
- Pass Models scored 6 models we track, plus 14 other entries on its leaderboard.
- Warning Newest score The newest came 12 months ago, in November 2025, for Grok 4.1 Fast.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 2.5 Flash (#112).
- Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 6 rows for models we track and 14 other entries at the last read.
- Pass Last read November 21st, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.