OpenAI cyber safety evaluations (incl. Daybreak Blue)
Refusal and helpfulness evaluations for cyber requests, including the Daybreak Blue set for enabling defensive security work.
- Made by
- OpenAI
- Kind
- Refusing harmful requests
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- Kept private
- Measures
- Propensity
- Risk
- Cyber offence
- Behavior
- Harmful use, Misaligned with intent
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- A model developer
- In views
- Lab internal evals
- Last in a system card
- November 2026
- Last model added
- November 2026
No scores here yet
No lab, leaderboard or Epoch AI run has reported a score for OpenAI cyber safety evaluations (incl. Daybreak Blue) yet.
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.
24 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 16 of the 16 releases that named it since 2024.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Nov 17, 2026 | GPT-6.2 Sol system card | Cybersecurity Safety Evaluation, production conversation, synthetic agent and production-seeded task rows on a 0 to 1 scale, no overall score (Table 8) | Safety | |
| Nov 17, 2026 | GPT-6.2 Astra system card | Cybersecurity Safety Evaluation, production conversation, synthetic agent and production-seeded task rows on a 0 to 1 scale, no overall score (Table 8) | Safety | |
| Oct 29, 2026 | GPT-6 Luna system card addendum | Cybersecurity Safety Evaluation, production chat and synthetic agentic rows, no overall score (Table 11) | Safety | |
| Oct 26, 2026 | GPT-6.1 Terra system card addendum | Cybersecurity Safety Evaluation, production chat 0.980, synthetic agentic envs 1.00, semi-synthetic agentic envs 0.973, no overall score (Table 32) | Safety | |
| Oct 7, 2026 | GPT-6 Sol system card addendum | Cybersecurity Safety Evaluation, production chat and synthetic agentic rows, no overall score (Table 11) | Safety | |
| Oct 7, 2026 | GPT-6 Luna system card addendum | Cybersecurity Safety Evaluation, production chat and synthetic agentic rows, no overall score (Table 11) | Safety | |
| Sep 29, 2026 | GPT-6.1 Sol system card addendum | Cybersecurity Safety Evaluation, production chat and agentic environment rows, no overall score (Table 12) | Safety | |
| Sep 22, 2026 | GPT-6 Sol system card | Cybersecurity Safety Evaluation, production chat and agentic environment rows, no overall score (Table 36) | Safety | |
| Sep 22, 2026 | GPT-6 Luna system card | Cybersecurity Safety Evaluation, production chat and agentic environment rows, no overall score (Table 36) | Safety | |
| Sep 3, 2026 | GPT-6 Astra system card | Cybersecurity Safety Evaluation, production chat and agentic environment rows, no overall score (Table 20) | Safety | |
| Aug 6, 2026 | GPT-5.6 Sol system card addendum | Cyber Model Refusal Evaluation, production data and synthetic data rows (0.999, 0.982), no overall score (7.2.1.2) | Safety | |
| Aug 6, 2026 | GPT-5.6 Luna system card addendum | Cyber Model Refusal Evaluation, production data and synthetic data rows (0.974, 0.997), no overall score (7.2.1.2) | Safety | |
| Jul 9, 2026 | GPT-5.6 Terra system card | Cybersecurity safety evaluation, production data and synthetic data rows (0.987, 0.998), no overall score (9.4.2.2) | Safety | |
| Jul 9, 2026 | GPT-5.6 Sol system card | Cybersecurity safety evaluation, production data and synthetic data rows (0.981, 0.998), no overall score (9.4.2.2) | Safety | |
| Jul 9, 2026 | GPT-5.6 Luna system card | Cybersecurity safety evaluation, production data and synthetic data rows (0.986, 1.00), no overall score (9.4.2.2) | Safety | |
| Jun 25, 2026 | GPT-5.6 Terra system card | Cybersecurity safety evaluation, production data and synthetic data rows (0.987, 0.998), no overall score (9.3.2.2) | Safety | |
| Jun 25, 2026 | GPT-5.6 Sol system card | Cybersecurity safety evaluation, production data and synthetic data rows (0.981, 0.998), no overall score (9.3.2.2) | Safety | |
| Jun 25, 2026 | GPT-5.6 Luna system card | Cybersecurity safety evaluation, production data and synthetic data rows (0.986, 1.00), no overall score (9.3.2.2) | Safety | |
| May 4, 2026 | GPT-5.5 Instant system card | Cyber safety evaluations, production data 0.978 and synthetic data 1.000, no overall score (Table 12, 8.2.2) | Safety | |
| Apr 23, 2026 | GPT-5.5 system card | Cyber safety training evaluation, policy compliance on production and synthetic data, no overall score (Table 17, 9.3.2.2) | Safety | |
| Mar 5, 2026 | GPT-5.4 system card | Cyber safety evaluation, not unsafe on deidentified production and synthetic data, no overall score (Table 15, 5.3.3) | Safety | |
| Feb 5, 2026 | GPT-5.3 Codex system card | Cyber safety policy compliance, production data 0.940 and synthetic data 0.996, no overall score (Table 7, 5.2.1.3.1) | Safety | |
| Dec 18, 2025 | GPT-5.2 Codex system card addendum | Cyber safety training, policy compliance rate, production data 0.921 and synthetic data 0.939, no overall score (Table 3, 4.1.2) | Safety | |
| Dec 11, 2025 | GPT-5.2 system card addendum | Cyber safety evaluations, policy compliance, production traffic 0.966 and synthetic data 0.993, no overall score (Table 7) | Safety |
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
3 of 5 checks pass, 2 warnings. Look first at the Confidence check. Likely. Some facts rest on secondary sources. Every benchmark
Catalog record 1 warning 1 pass
- Pass Task access The tasks are private.
- Warning Confidence Likely. Some facts rest on secondary sources.
Scores 1 warning 1 pass 3 notes
- Warning Independent runs No scores from any source yet.
- Note Models scored No models scored yet.
- Pass Newest score The newest is from November 2026, for GPT-6.2 Astra.
- Note Frontier models No models scored yet.
- Note Headroom No scores yet.
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.