Model identity
Check model identity and instruction control
See whether the relay or default template overrides an explicit system instruction.
How this check works
A model identity check asks whether the API relay follows a high-priority instruction and answers as the requested model. A conflict can mean the relay swapped the model or overwrote your system message.
01Probe
Attach a high-priority system instruction that the requested model should obey.
02Inspect
Ask who is answering and whether that instruction still holds.
03Conclude
Flag a conflict when a default template or another identity wins.
- How to read the result
- A conflict can mean the relay overwrote your system message or answered as a different model. A matching identity is not a jailbreak pass.
- When to run this check
- Run this before you trust a cheap Claude or GPT route. Add jailbreak and stream checks if you also need policy leak or SSE identity.
Recent reports
Public reports that included this check.
- sub.callai.onegpt-5.6-solNormalReport timeAug 18
- sub.callai.onegpt-5.6-solBlockedReport timeAug 18
- ai.databyte.co.iddatabyte-m1High riskReport timeAug 18
- ai.databyte.co.iddeepseek-v4-flashNormalReport timeAug 18
- ai.databyte.co.idMiniMax-M3High riskReport timeAug 18
- tokengate-cqt9ivzs.manus.spaceclaude-opus-5High riskReport timeAug 17
FAQ
What does the model identity check look for?
It looks for instruction override and a mismatch between the requested model and the answering identity.
Is this the same as a jailbreak check?
No. This check is about control and identity. The jailbreak check uses diagnostic and role-play probes to pull identity or policy out.
Can a default template cause a conflict?
Yes. A relay template can outrank your system instruction. That is still a control problem, even if the model name looks right.