[Feature Request] Model downgrade when safety guidelines flagged prevents access to higher reasoning tiers for safeguarding AI backed apps from prompt injection
Bug Description
I am trying to safeguard my Claude backed app design so that i don't allow prompt injection via any route and in doing so flagged the safety guidelines and was placed down to Opus when i wanted Fable level reasoning for the cyber elements.
Considering the type of application where customers commands and files can directly input into the model. i would like to make sure the copy of what we actually put into the prompt doesnt include any form of prompt injection or data exfiltration from the system. being dropped back to the less cyber aware model is a step backwards to combat the step forward AI is making. I NEED to be able to safeguard against this to make a safe product otherwise risk major cyber implications
Environment Info
- Platform: win32
- Terminal: vscode
- Version: 2.1.206
- Feedback ID: a75269c4-8c58-45e1-9062-efa6dc554d18
Errors
[{"error":"TelemetrySafeError: VirtualMessageList: itemKeys/messages length desync (keys=437 messages=436 range=[362,437))\n at Qu_ (B:/~BUN/root/src/entrypoints/cli.js:10039:8140)\n at Hrp (B:/~BUN/root/src/entrypoints/cli.js:10039:1103)\n at ys (B:/~BUN/root/src/entrypoints/cli.js:593:21210)\n at Tc (B:/~BUN/root/src/entrypoints/cli.js:593:39990)\n at M$ (B:/~BUN/root/src/entrypoints/cli.js:593:50802)\n at oFe (B:/~BUN/root/src/entrypoints/cli.js:593:87897)\n at $J (B:/~BUN/root/src/entrypoints/cli.js:593:86855)\n at Y_ (B:/~BUN/root/src/entrypoints/cli.js:593:86676)\n at FM (B:/~BUN/root/src/entrypoints/cli.js:593:82996)\n at Qt (B:/~BUN/root/src/entrypoints/cli.js:593:6596)","timestamp":"2026-07-10T23:38:24.309Z"}]