Model
operator
Appears alongside
Featured in threads
Tracks
- Safety & alignment 2
- Security & misuse 1
- Models & capabilities 1
OpenAI publishes 'Security on the path to AGI'
OpenAI described using its own models for threat detection, hired SpecterOps for continuous red-teaming, and said it was building prompt-injection defences into its Operator agent.
Security & misuse · Safety & alignment
OpenAI launches Operator
Built on a new Computer-Using Agent model layered on GPT-4o, it scored 38.1% on OSWorld against a 72.4% human baseline, and launched to $200-a-month Pro subscribers only.
Models & capabilities
OpenAI publishes Operator system card
External red-teamers targeted prompt injection specifically; OpenAI reported raising its injection-detection recall from 79% to 99% after one testing round.
Safety & alignment