[Bug] Verified Cyber Program account still experiencing model downgrade on authorized security work

Bug Description
Hi,

I was accepted into the Cyber Verification Program on 2026-10-05. My understanding is that verified accounts get fewer
cyber-safeguard interruptions on authorized security work.

I'm doing authorized AI/LLM red-teaming for Mozilla's 0DIN GenAI bug bounty (human-in-the-loop, synthetic canaries only,
no real payloads). In Claude Code, requests that load my red-team workflow are being classified as cyber and silently
stepped down from Opus to a smaller model, which is the behavior I expected verification to relax.

Two questions:

1. Is the downgrade expected to still apply to verified accounts, or should verification have relaxed it? If it should

have, has my verification fully propagated to the account tied to akrose858@gmail.com?

1. If some downgrade is expected even for verified users, how do you recommend verified researchers structure this work

so authorized testing runs on the intended model?

I want to stay inside the sanctioned path, so I'd rather get your guidance than work around anything. Happy to share
specifics of the engagement.

Thanks,
Anthony Rose

Environment Info

• Platform: win32
• Terminal: null
• Version: 2.1.289
• Feedback ID: 6d63c412-0ff3-4db6-b11a-df46a8deb3b8