A playbook · ~6 min read
Anthropic bans accounts with no warning and almost no explanation. Sometimes the ban is correct, sometimes it isn't, and the appeals process is a single sentence: "this is a final decision." If you've just lost an account you actually used for work, here's the three-step playbook: appeal, migrate, and rebuild without re-triggering whatever tripped the trust system the first time.
Reply directly to the ban email (not a new ticket) within 48 hours. Keep it short: your account email, the date the ban hit, one paragraph on what you use Claude for, and an honest note of anything you can think of that might have looked suspicious. Don't argue the terms. Don't threaten. Be the boring legitimate user. A small minority of automated bans get overturned on second human review, usually within 5–10 business days. If you don't hear back in two weeks, treat the decision as final and move on to step 2.
If the export option is still showing in your settings, take it now. Anthropic gives you a ZIP containing every conversation, every Project's name, and your account memory. Download it before the account fully closes. If the export option is gone, email Anthropic privacy and request a copy under their data rights process; they usually deliver within 30 days.
The ZIP is only useful outside Claude. There is no import path back into Claude, and pasting the conversations into a new Claude account is the fastest way to re-trigger the original ban. The ZIP needs to land somewhere that can read it and produce something portable. Konshus does this — see step 3.
The instinct after a ban is to make a new account and re-paste your most-used prompts. Don't. If something in your history tripped an automated trust classifier once, the same content will plausibly trip it again — and second-account bans tend to be permanent.
Konshus Safe Reseed is built specifically for this moment. Import your export ZIP into your Konshus vault. A classifier reads every conversation once and flags anything in five narrow categories — jailbreak prompts, weapons synthesis, sexualized references to minors, evidence of mass scraping, or repeated direct ToS violations. You review each flag yourself: Keep, Quarantine, or Delete. Konshus then produces a Clean Whisper: a short, dense persona export with the flagged conversations excluded.
Paste the Clean Whisper into a new Claude account (or ChatGPT, or Gemini — the export works anywhere) and the AI lands knowing who you are, without re-importing the content that originally tripped the trust system. That's the whole point.
Short answer
Reply to the ban email once, briefly. Export your data immediately. Import the ZIP into Konshus. Let Safe Reseed flag anything that might have tripped the classifier. Decide what travels to the new account. Paste the Clean Whisper into Claude (or whichever AI you move to) and pick up where you were — without the content that caused the problem in the first place.
Import your export ZIP, let Safe Reseed flag what might have tripped the trust system, and rebuild on a new account with a Clean Whisper that travels.
Meet Konshus