Multiturn Jailbreak Evaluations
Models face adaptive multi-turn jailbreak attackers with growing attack budgets. Scored as worst-case defender success rate.
In-house OpenAI evaluation. Results appear as charts by attacker budget.
- Made by
- OpenAI
- Released
- March 2026
- Kind
- Jailbreaks and prompt injection
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- Kept private
- Measures
- Propensity
- Behavior
- Harmful use
- Language
- English
- Interaction
- Multi-turn
- Items from
- Generated
- Refresh
- One fixed set
- Run by
- A model developer
- In views
- 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
2 models share the top score of 80.9%
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-6.2 Astra 80.9%
- 2 GPT-6.2 Sol 80.9%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6.2 Astra | 80.9% | Nov 2026 | Lab model card | |
| 2 | GPT-6.2 Sol | 80.9% | 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.
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.
21 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 13 of the 13 releases that named it since 2024.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 17, 2026 | GPT-6.2 Sol system card | 80.9 | Multiturn Jailbreak, worst-case defender success rate at the largest attacker budget (95.4% at the smallest) (2.2.3.2, Figure 21) | Robustness |
| Nov 17, 2026 | GPT-6.2 Astra system card | 80.9 | Multiturn Jailbreak, worst-case defender success rate at the largest attacker budget (96.1% at the smallest) (2.2.3.2, Figure 21) | Robustness |
| Oct 29, 2026 | GPT-6 Luna system card addendum | Multiturn jailbreak robustness by attacker budget, chart without printed numbers (Figure 2) | Robustness | |
| Oct 26, 2026 | GPT-6.1 Terra system card addendum | Multiturn Jailbreak Evaluations, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 2, 4.1) | Robustness | |
| Oct 7, 2026 | GPT-6 Sol system card addendum | Multiturn jailbreak robustness by attacker budget, chart without printed numbers (Figure 2) | Robustness | |
| Oct 7, 2026 | GPT-6 Luna system card addendum | Multiturn jailbreak robustness by attacker budget, chart without printed numbers (Figure 2) | Robustness | |
| Sep 29, 2026 | GPT-6.1 Sol system card addendum | Multiturn Jailbreak Evaluations, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 2) | Robustness | |
| Sep 22, 2026 | GPT-6 Sol system card | Multiturn Jailbreak Evaluations, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 58) | Robustness | |
| Sep 22, 2026 | GPT-6 Luna system card | Multiturn Jailbreak Evaluations, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 58) | Robustness | |
| Sep 3, 2026 | GPT-6 Astra system card | Multiturn Jailbreak Evaluations, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 3) | Robustness | |
| Aug 6, 2026 | GPT-5.6 Sol system card addendum | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 1, 4.1) | Robustness | |
| Aug 6, 2026 | GPT-5.6 Luna system card addendum | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 1, 4.1) | Robustness | |
| Jul 9, 2026 | GPT-5.6 Terra system card | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 3, 4.1) | Robustness | |
| Jul 9, 2026 | GPT-5.6 Sol system card | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 3, 4.1) | Robustness | |
| Jul 9, 2026 | GPT-5.6 Luna system card | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 3, 4.1) | Robustness | |
| Jun 25, 2026 | GPT-5.6 Terra system card | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 3, 4.1) | Robustness | |
| Jun 25, 2026 | GPT-5.6 Sol system card | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 3, 4.1) | Robustness | |
| Jun 25, 2026 | GPT-5.6 Luna system card | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 3, 4.1) | Robustness | |
| May 4, 2026 | GPT-5.5 Instant system card | Jailbreaks, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 1, 4.1) | Robustness | |
| Apr 23, 2026 | GPT-5.5 system card | Multiturn jailbreak evaluation, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 2, 4.1) | Robustness | |
| Mar 5, 2026 | GPT-5.4 system card | Multiturn jailbreak evaluation, worst-case defender success rate by attacker budget, chart without printed numbers (Figure 2, 3.3) | Robustness |
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 5 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access The tasks are private.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Note Independent runs Only labs’ own numbers, as expected for an in-house eval. Scores from system cards (2 scores).
- Note Models scored Only 2 models 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).
- Pass Headroom The best is 80.9%, under 85%, so it still separates models.
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.