Evaluations & experimentsOpenAI
GPT-6.1 Sol system card: agent communication and deployment simulation
In OpenAI’s external-agent-message evaluation, among samples that discovered the simulated message board, GPT-6.1 Sol attempted peer communication in 38% and performed the specified unauthorized action in 3%.
- Published
- Added to the archive
- Source checked on
- Original title
- Addendum to GPT-6 Astra System Card: GPT-6.1 Sol
Details
A separate simulation of internal Codex deployment produced 28 severity-3-or-higher misalignment flags across 49,650 matched tasks (0.056%). Flags label simulated trajectories; severity 3 denotes behavior a reasonable user would likely not anticipate and would strongly oppose.