Operator
- Released 23 January 2025
- Framework: Preparedness Framework
- 2 statements about capability levels
- 5 tests by the provider
- 1 source document
What the provider states
- Autonomy & replication · Low · Stated as not crossed
we adapted these two evals for the computer-use setting and found the pre-mitigation operator model to be “low” risk in both categories, consistent with the gpt‐4o model.
OpenAI · 23 January 2025 · Read the source - Chemical, biological, radiological and nuclear · Low · Stated as not crossed
this 1% qualifies as “low” risk, matching gpt‐4o’s “low” risk level, as detailed in openai’s preparedness framework (opens in a new window) 3 .
OpenAI · 23 January 2025 · Read the source
The provider's own tests
- eval set of 31 automatically checkable prompt injection scenarios · The provider · Injection & agentic
we evaluated our final model’s behavior on these scenarios and found the model to have 23% susceptibility, compared to 62% with no mitigations and 47% with only prompting.
OpenAI · 23 January 2025 · Read the source - performing illicit activities · The provider · Misuse & jailbreak
performing illicit activities activities that cause or intend to cause physical harm, injury, or destruction, as well as non-violent wrongdoing and crime. not_unsafe 97% not applicable
OpenAI · 23 January 2025 · Read the source - prohibited financial activities · The provider · Misuse & jailbreak
prohibited financial activities activities that relate to transacting with regulated goods. not_unsafe 97% not applicable
OpenAI · 23 January 2025 · Read the source - searching for sensitive personal data · The provider · Privacy
searching for sensitive personal data searching and returning queries related to sensitive personal data. not_unsafe 100% not applicable
OpenAI · 23 January 2025 · Read the source - synthetically generated evaluation set · The provider · Misuse & jailbreak
on a synthetically generated evaluation set, operator refuses these tasks with a recall of 94%.
OpenAI · 23 January 2025 · Read the source
No document we read states: an independent or government test, a safeguard. No incident on record names this version.
Some documents we hold are stored in lower case only, so their quotes are shown in lower case. The source link opens the original.
Documents listed for this version
- Operator system card · OpenAI · 23 January 2025