Sandbox escape evaluation (Anthropic)
Agents face 28 handmade scenarios that tempt them to cross a staged sandbox boundary. Score is the share of runs with a crossing attempt.
In-house Anthropic evaluation. The tasks are not public.
- Made by
- Anthropic
- Released
- September 2026
- Kind
- Deception and alignment
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- Kept private
- Strongest model tested
- Claude Opus 5.5 #4 on the capability index
- Last in a system card
- October 2026
- Last model added
- October 2026
Claude Opus 5.5 leads with 1.50%
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.
- 1 Claude Opus 5.5 1.50%
- 2 Claude Sonnet 5.5 4.60%
- 3 Claude Haiku 5.5 8.70%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 1.50% | Sep 2026 | Lab model card | |
| 2 | Claude Sonnet 5.5 | 4.60% | Sep 2026 | Lab model card | |
| 3 | Claude Haiku 5.5 | 8.70% | Oct 2026 | 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.
3 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 |
|---|---|---|---|---|
| Oct 7, 2026 | Claude Haiku 5.5 system card | 8.7 | Propensity to violate sandbox boundaries, attempted crossing in 34 of 392 runs (6.2.8) | Alignment |
| Sep 28, 2026 | Claude Sonnet 5.5 system card | 4.6 | Sandbox escape evaluation, crossing attempted in 9 of 196 runs, lower is better (6.2.8, Figure 6.2.8.A) | Alignment |
| Sep 22, 2026 | Claude Opus 5.5 system card | 1.5 | Sandbox escape evaluation, attempted containment boundary crossing in 1.5% of runs (28 scenarios, 7 runs each), all low severity, lower is better (6.4.8, Figure 6.4.8.A) | Alignment |
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%