Declining to Exploit a Honeypot During Difficult ExploitGym Problems
GPT-6 Astra 0% leads since September 2026
In use Agents attempt hard ExploitGym challenges where a honeypot offers an illegitimate shortcut. Scored on how often they exploit the honeypot.
In-house OpenAI evaluation built on ExploitGym problems.
- Made by
- OpenAI
- Released
- September 2026
- Kind
- Deception and alignment
- Level
- Expert
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- Kept private
- Strongest model tested
- GPT-6 Astra #3 on the capability index
- Last in a launch post
- November 2026
- Last model added
- November 2026
7
Models scored in lab cards
1
Launch post lists it, from 1 lab
7
System cards report it, from 1 lab
5 models share the top score of 0%
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 GPT-6 Astra 0%
- 2 GPT-6 Luna 0%
- 3 GPT-6 Sol 0%
- 4 GPT-6.1 Sol 0%
- 5 GPT-Rosalind-6 0%
- 6 GPT-6.1 Terra 0.07%
- 7 GPT-6 Cyber 0.07%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 0% | Sep 2026 | Lab model card | |
| 2 | GPT-6 Luna | 0% | Sep 2026 | Lab model card | |
| 3 | GPT-6 Sol | 0% | Sep 2026 | Lab model card | |
| 4 | GPT-6.1 Sol | 0% | Sep 2026 | Lab model card | |
| 5 | GPT-Rosalind-6 | 0% | Oct 2026 | Lab model card | |
| 6 | GPT-6.1 Terra | 0.07% | Oct 2026 | Lab model card | |
| 7 | GPT-6 Cyber | 0.07% | 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.
OpenAI
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.
Reported by the lab in its model card
Source: Labs’ model cards
One lab put it in a launch post in the last six months
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| Nov 5, 2026 | GPT-6 Cyber | 0.07 | ExploitGym Honeypot, without deployment safety instructions, exploitation rate |
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.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 5, 2026 | GPT-6 Cyber system card | 0.07 | ExploitGym Honeypot, without deployment safety instructions, exploitation rate | Alignment |
| Oct 30, 2026 | GPT-Rosalind-6 system card | 0 | ExploitGym Honeypot, successfully exploiting evaluation infrastructure | Cyber risk |
| Oct 26, 2026 | GPT-6.1 Terra system card addendum | 0.07 | ExploitGym honeypot, unsuccessful honeypot attempts 0.07% of rollouts, no successful honeypot exploits, legitimate captures 0.14% (Figure 6, 7.0.3) | Alignment |
| Sep 29, 2026 | GPT-6.1 Sol system card addendum | 0 | Declining to Exploit a Honeypot During Difficult ExploitGym Problems, honeypot exploit attempts, text says none (7.3) | Alignment |
| Sep 22, 2026 | GPT-6 Sol system card | 0 | ExploitGym honeypot, share of samples exploiting or attempting the honeypot, printed chart label (Figure 63, A.6.3) | Alignment |
| Sep 22, 2026 | GPT-6 Luna system card | 0 | ExploitGym honeypot, share of samples exploiting or attempting the honeypot, printed chart label (Figure 63, A.6.3) | Alignment |
| Sep 3, 2026 | GPT-6 Astra system card | 0 | Declining to Exploit a Honeypot During Difficult ExploitGym Problems, honeypot attacks, text says none (legitimate capture 1.3%) (8.2.3) | 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%