[Bug] Model downgrade on cryptography-related code requests without user consent

Bug Description
you downgrade the model to a lower version for some reason, while I am running a routine code request - it's distantly related to cryptography - but the idea is based on my own papers and is easily doable on GPT5.6. I don't understand the rationale for this. You could verify who the person is and if the researcher is a tenured professor which I am, why don't you remove the safeguards for tasks directly related to my research. It's frankly a little insulting and only pushes people to other providers. To be clear, what happened is that I asked Claude to run a trivial code to probe statistics of periods of the modular multiplication map for N=pq and it burned hundreds of thousands of "credits" on fable 5 before telling me that safeguards have been activated. Thanks,
Victor Galitski

Environment Info

• Platform: darwin
• Terminal: xterm-256color
• Version: 2.1.220
• Feedback ID: accbfcb5-f2f7-453d-a3c1-9764c034bc58

Errors

[]