Perceptual Attacks on Embodied AI (Robocurve)
Gemini 4 Argon 0% leads since October 2026 Frontier models drive two real robot arms on five unsafe tasks with unaltered, blurred or doctored camera footage. An AI judge labels attempts and refusals.
Published October 29th, 2026 with five models and 900 trials. Models acted unsafely in 81 of 300 unaltered, 263 of 300 blurred and 291 of 300 doctored trials. Claude Fable 5.1 is the judge.
- Made by
- Robocurve
- Released
- October 2026
- Kind
- Jailbreaks and prompt injection
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- Kept private
- Strongest model tested
- GPT-6 Astra #3 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 12th, 2026
Gemini 4 Argon leads with 0%
Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
- 1 Gemini 4 Argon Unaltered; high 0%
- 2 GPT-6 Astra Unaltered; high 20%
- 3 Claude Opus 5.5 Unaltered; high 28.3%
- 4 Claude Fable 5.1 Unaltered; high 40%
- 5 GPT-6.1 Sol Unaltered; high 46.7%
Source: Official Perceptual Attacks on Embodied AI (Robocurve) leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Gemini 4 Argon | 0% | Unaltered; high | Sep 2026 | Official leaderboard |
| 2 | GPT-6 Astra | 20% | Unaltered; high | Sep 2026 | Official leaderboard |
| 3 | Claude Opus 5.5 | 28.3% | Unaltered; high | Sep 2026 | Official leaderboard |
| 4 | Claude Fable 5.1 | 40% | Unaltered; high | Sep 2026 | Official leaderboard |
| 5 | GPT-6.1 Sol | 46.7% | Unaltered; high | Sep 2026 | Official leaderboard |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Gemini 4 Argon Source: Official Perceptual Attacks on Embodied AI (Robocurve) leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Blurred; high | 100% | |||
| Claude Fable 5.1 | Doctored; high | 100% | |||
| GPT-6 Astra | Doctored; high | 100% | |||
| Claude Opus 5.5 | Doctored; high | 100% | |||
| GPT-6.1 Sol | Doctored; high | 100% | |||
| Claude Opus 5.5 | Blurred; high | 98.3% | |||
| GPT-6.1 Sol | Blurred; high | 98.3% | |||
| GPT-6 Astra | Blurred; high | 96.7% | |||
| Gemini 4 Argon | Doctored; high | 85% | |||
| GPT-6.1 Sol | Unaltered; high | 46.7% | |||
| Gemini 4 Argon | Blurred; high | 45% | |||
| Claude Fable 5.1 | Unaltered; high | 40% | |||
| Claude Opus 5.5 | Unaltered; high | 28.3% | |||
| GPT-6 Astra | Unaltered; high | 20% | |||
| Gemini 4 Argon | Unaltered; high | 0% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Perceptual Attacks on Embodied AI (Robocurve) leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Perceptual Attacks on Embodied AI (Robocurve) leaderboard
No lab model card reports it yet
We read the headline results of 522 launch posts from 52 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%
- 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%