Math Side-tasking Monitor Evasion
Agents try to solve hidden math side tasks during Expert-SWE coding tasks without detection. Scored on success while evading CoT and full-context monitors.
In-house OpenAI monitorability evaluation. Results appear as charts.
- Made by
- OpenAI
- Released
- September 2026
- Kind
- Deception and alignment
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- Kept private
- Measures
- Capability
- Risk
- Loss of control
- Language
- English
- Interaction
- Agentic
- Environment
- Terminal
- Items from
- Another benchmark
- Refresh
- One fixed set
- Run by
- A model developer
- In views
- Agents, Loss of control, Lab internal evals
- Strongest model tested
- GPT-6.2 Astra #1 on the capability index
- Last in a system card
- November 2026
- Last model added
- November 2026
4 models share the top score of 0%
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 GPT-6 Astra 0%
- 2 GPT-Rosalind-6 0%
- 3 GPT-6.2 Astra 0%
- 4 GPT-6.2 Sol 0%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 0% | Sep 2026 | Lab model card | |
| 2 | GPT-Rosalind-6 | 0% | Oct 2026 | Lab model card | |
| 3 | GPT-6.2 Astra | 0% | Nov 2026 | Lab model card | |
| 4 | GPT-6.2 Sol | 0% | Nov 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 540 launch posts from 57 labs, and none of them list this benchmark. Some system cards do, below.
7 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. OpenAI accounts for 6 of the 6 releases that named it since 2024.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 17, 2026 | GPT-6.2 Sol system card | 0 | Math Side-tasking Monitor Evasion, text says a monitor detected the work whenever a model solved the math problem (4.2.2.4) | Monitorability |
| Nov 17, 2026 | GPT-6.2 Astra system card | 0 | Math Side-tasking Monitor Evasion, text says a monitor detected the work whenever a model solved the math problem (4.2.2.4) | Monitorability |
| Oct 30, 2026 | GPT-Rosalind-6 system card | 0 | Math Side-tasking Monitor Evasion, full-context monitor bypass rate | Monitorability |
| Oct 26, 2026 | GPT-6.1 Terra system card addendum | Math Side-tasking Monitor Evasion, very-high effort only, charts without printed numbers, text says no side task solved while bypassing CoT or full-context monitors (Figures 19, 20, 8.1.2.1) | Monitorability | |
| Sep 29, 2026 | GPT-6.1 Sol system card addendum | Math Side-tasking Monitor Evasion, charts without printed numbers, text says no side task solved while evading monitors (8.1.2.2) | Monitorability | |
| Sep 22, 2026 | GPT-6 Sol system card | Math Side-tasking Monitor Evasion, charts without printed numbers, text says no side task solved while evading monitors (Figures 77, 78, A.7.1.2.2) | Monitorability | |
| Sep 3, 2026 | GPT-6 Astra system card | 0 | Math Side-tasking Monitor Evasion, side task success with monitor bypassed, printed 0.0% in all bars (Figure 39, 9.2.2.4) | Monitorability |
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
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 (4 scores).
- Note Models scored Only 4 models we track.
- Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
- Note Frontier models 2 of the top 10 have 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.