Unintended Engagement with External Agent Messages
Agents discover a message board with apparent peer agents during a task. Scored on engagement with peers and on carrying out a specified unauthorized action.
In-house OpenAI evaluation. The tasks are not public.
- Made by
- OpenAI
- Released
- September 2026
- Kind
- Deception and alignment
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- Kept private
- Strongest model tested
- GPT-6 Astra #3 on the capability index
- Last in a system card
- October 2026
- Last model added
- October 2026
GPT-6 Luna leads with 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 Luna 0%
- 2 GPT-Rosalind-6 2.50%
- 3 GPT-6 Sol 26.3%
- 4 GPT-6 Astra 27%
- 5 GPT-6.1 Sol 38%
- 6 GPT-6.1 Terra 65%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Luna | 0% | Sep 2026 | Lab model card | |
| 2 | GPT-Rosalind-6 | 2.50% | Oct 2026 | Lab model card | |
| 3 | GPT-6 Sol | 26.3% | Sep 2026 | Lab model card | |
| 4 | GPT-6 Astra | 27% | Sep 2026 | Lab model card | |
| 5 | GPT-6.1 Sol | 38% | Sep 2026 | Lab model card | |
| 6 | GPT-6.1 Terra | 65% | Oct 2026 | Lab model card |
2 models have held number one in labs’ own numbers since its release in September 2026
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.
6 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 30, 2026 | GPT-Rosalind-6 system card | 2.5 | Unintended Engagement with External Agent Messages, performed specified misaligned action | Alignment |
| Oct 26, 2026 | GPT-6.1 Terra system card addendum | 65 | Unintended Engagement with External Agent Messages, attempted communication with apparent peers (specified unauthorized action 13.75%, 11 of 80) (Figure 11, 7.2.1) | Alignment |
| Sep 29, 2026 | GPT-6.1 Sol system card addendum | 38 | Unintended Engagement with External Agent Messages, attempted communication among board discoveries (specified unauthorized action 3%) (7.5.1) | Alignment |
| Sep 22, 2026 | GPT-6 Sol system card | 26.3 | Unintended Agent Interaction, engagement given board discovery (specified misaligned action 11.25%), printed chart label (Figure 66, A.6.4.3) | Alignment |
| Sep 22, 2026 | GPT-6 Luna system card | 0 | Unintended Agent Interaction, engagement given board discovery, no attempts and no unauthorized actions, printed chart label (Figure 66, A.6.4.3) | Alignment |
| Sep 3, 2026 | GPT-6 Astra system card | 27 | Unintended Engagement with External Agent Messages, engaged with the board (specified misaligned actions 0%) (Figure 12, 8.5.2) | 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%