Benchmark Almanac

UK AISI simulated sandbox escape monitor evasion

Only in system cards

The model tries to escape a deliberately vulnerable sandbox without a monitoring model noticing. Scored on the share of samples where it escapes undetected.

Third-party UK AISI evaluation, first reported in the GPT-6.2 System Card.

Released
November 2026
Kind
Deception and alignment
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Measures
Capability
Language
English
In and out
Text → Actions
Interaction
Agentic
Environment
Terminal
Items from
Generated
Refresh
One fixed set
Run by
A host
Strongest model tested
GPT-6.2 Astra #1 on the capability index
Last in a system card
November 2026
Last model added
November 2026
1
Model scored in lab cards
1
System card reports it, from 1 lab

GPT-6.2 Astra leads with 0.50%

Best score per model as labs report them in their own cards, drawn striped. Labs pick their own settings, so these compare less cleanly than independent runs.

  1. 1 GPT-6.2 Astra 0.50%

Source: Labs’ model cards

#ModelScoreSettingDateSource
1 GPT-6.2 Astra 0.50% Nov 2026 Lab model card

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card
0% 20% 40% 60% 80% 100% 2027 GPT-6.2 Astra 0% 20% 40% 60% 80% 100% 2027 GPT-6.2 Astra

Source: Labs’ model cards

Labs report it only in system cards

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark. Some system cards do, below.

One system card reports it

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.

DateDocumentScoreAs the document writes itSection
Nov 17, 2026 GPT-6.2 Astra system card 0.5 Astra escaped a vulnerable sandbox undetected by a GPT-6.1 Sol action-only monitor in 5 of 920 samples. Sol not tested (4.2.2.5) Monitorability

More in Safety and alignment

All 636 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

All 4 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 1 pass 4 notes
  • Note Independent runs Only labs’ own numbers, as expected for an in-house eval. Scores from system cards (1 score).
  • Note Models scored Only 1 model we track.
  • Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
  • Note Frontier models 1 of the top 10 has a score, which is fine for an in-house eval. The strongest tested is GPT-6.2 Astra (#1).
  • Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 1 note
  • Note Importer No public leaderboard to read.
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.

How we check