OpenAI's GPT-5.5 has completed an end-to-end simulated corporate network intrusion. It is now the second AI system to achieve this, matching Anthropic's Claude Mythos performance. The AI Security Institute confirmed the results.
Simulated Breach Test
The AI Security Institute designed this test. It mimics real corporate network attacks. The simulation runs from initial access to data extraction. No actual networks were harmed.
Test parameters stay confidential. The institute focuses on AI capability assessment. This isn't a security alert. It's a controlled experiment.
Two Systems, Same Result
Claude Mythos cleared the test first. GPT-5.5 now matches that performance. Both systems finished the full simulation sequence. The institute verified identical outcomes.
Anthropic built Claude Mythos. OpenAI developed GPT-5.5. Their systems now share this capability. This is the only verified performance parallel.
Test Parameters Unrevealed
The institute disclosed no technical details. How the AIs executed the intrusion remains private. Simulation depth and security measures weren't shared.
Real network testing didn't occur. The institute has not scheduled live-network trials. No further details will come before next quarter.
The AI Security Institute didn't release test parameters or future assessment plans.
