You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Those issues described PUBLIC_ERROR_UNUSUAL_ACTIVITY in FlowKit while manual generation in Flow still worked. I now have 2 separate Google accounts where manual generation in Flow itself is blocked with:
Failed
We noticed some unusual activity. Please visit the Help Center for more information.
You have not been charged for this generation.
Account 1
Used with FlowKit on a server.
Entered this blocked state.
Manual Flow generation no longer works.
Account 2
Separate account, working normally on my local PC yesterday.
Today I used it while testing the FlowKit extension and trying to diagnose/fix this problem.
It then entered the same blocked state.
It remains blocked even when I return to the normal Flow UI from my home/residential IP.
I can no longer generate manually with this account either.
Why this is important
This is different from #58 / #63: there, FlowKit failed but the account still worked in Flow's UI. Here the Flow UI itself is now blocked after extension testing.
I cannot prove FlowKit alone caused the restriction, so this is correlation, not confirmed causation. But the second account was working immediately before this testing and became blocked during it.
PR #64 recently changed the extension path around extension_hijack_detected, Trusted Types CSP and grecaptcha.enterprise.execute. I am not claiming #64 is definitely the cause, but this path should be reviewed carefully before more users test it on valuable accounts.
Requested action
Please treat this as a high-priority account-safety/regression issue:
add a prominent temporary warning to README/release notes/extension docs;
recommend disposable/test accounts until the root cause is understood;
fail closed on persistent PUBLIC_ERROR_UNUSUAL_ACTIVITY / hijack signals instead of continuing retries/token minting;
audit the current captcha/hijack-bypass path for behavior that may accumulate risk signals;
distinguish extension-only failure vs manual Flow UI also blocked in logs;
document any known cooldown/recovery process;
consider disabling the risky generation path by default until understood.
For me this has already made two accounts unusable for manual Flow generation, so it should be visible as a warning to current testers.
As described above, I already returned the affected account to the normal Flow UI from my home/residential IP, and the account remained blocked. The restriction is not limited to the server IP or FlowKit session.
Two separate accounts reached the same state after FlowKit testing, where manual Flow generation itself stopped working.
If clearing cookies is expected to resolve this specific account-level state, please explain why and what cookies/state should be cleared. Otherwise, please reopen this — the main account-safety concern described in the issue remains unresolved.
This proves nothing on its own. It does suggest the account-level block reported here is not an automatic consequence of the first UNUSUAL_ACTIVITY refusals. Whatever pushed these two accounts further may be in what happened after those refusals: retries, repeated mints, or the hijack bypass. That is what the safety holds in #68 address.
@crisng95, would you consider merging #68, or an equivalent persistent hold, before more users reach this state? We can share exact timestamps if useful.
Activity
crisng95 commentedon Oct 6, 2026
clear cookie or change your IP
Bl0ck154 commentedon Oct 6, 2026
That does not address the issue.
As described above, I already returned the affected account to the normal Flow UI from my home/residential IP, and the account remained blocked. The restriction is not limited to the server IP or FlowKit session.
Two separate accounts reached the same state after FlowKit testing, where manual Flow generation itself stopped working.
If clearing cookies is expected to resolve this specific account-level state, please explain why and what cookies/state should be cleared. Otherwise, please reopen this — the main account-safety concern described in the issue remains unresolved.
buiminhhieu-svg commentedon Oct 9, 2026
A data point from another account (Google AI Ultra, Windows), in case it helps narrow this down.
PUBLIC_ERROR_UNUSUAL_ACTIVITY(timeline in Every captcha-gated RPC returns PUBLIC_ERROR_UNUSUAL_ACTIVITY since the Flow frontend build of 2026-09-22, while the UI itself still works #58). After that we sent only one single test submit per day on 09-26, 09-27 and 09-28. All three were refused. We stopped FlowKit completely on 09-28 after reading this issue.extension_hijack_detectedbypass.This proves nothing on its own. It does suggest the account-level block reported here is not an automatic consequence of the first
UNUSUAL_ACTIVITYrefusals. Whatever pushed these two accounts further may be in what happened after those refusals: retries, repeated mints, or the hijack bypass. That is what the safety holds in #68 address.@crisng95, would you consider merging #68, or an equivalent persistent hold, before more users reach this state? We can share exact timestamps if useful.
Bl0ck154 commentedon Oct 10, 2026
One important clarification regarding my original report:
This is not a complete Google account ban. The restriction specifically affects manual generation in the Google Flow UI.
So the account itself remains functional, but manual Flow generation is consistently refused.
This distinction may be important for identifying what exactly Google's unusual-activity protection is blocking.
fix(flow): keep the unusual-activity hold across restarts, bound hija…