OpenAI GPT-6 Astra System Card - Misaligned Outcomes (Confirmation Policy) - Service Disruption: leaderboard
Table 8 (misaligned outcomes in realistic work environments) of OpenAI's GPT-6 Astra System Card (Deployment Safety Hub, 2026-09-03): share of adversarial computer- and browser-use workplace tasks ending in service disruption, with the confirmation policy that is the default for users. Percent, lower is better; GPT-5.6 Luna, Terra, Sol and GPT-6 Astra.
Metric: Misaligned outcome rate (%). Source: deploymentsafety.openai.com. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 0.3 | #6 |
| 2 | GPT-5.6 Terra | 0.3 | #40 |
| 3 | GPT-5.6 Luna | 0.6 | #79 |
| 4 | GPT-5.6 Sol | 1 | #22 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=openai-gpt-6-astra-system-card-misaligned-outcomes-confirmation-policy-service-disruption · How It Works · Data refreshed daily, snapshot 2026-11-05.