HuaRenCa
Back to Forum
Community

Fable keeps triggering safety flag Frontier_llm

大航海
大航海

2 weeks ago

As the title says, lately Fable has been triggering the "frontier_llm" flag every day, saying I'm working on a competing LLM and then blocking me from using it. Does anyone have experience with this?

I really can't get it to work, so I'm thinking of wiping the cc leftovers on the machine, reinstalling and switching accounts, then changing to Astra Harness Fable+5.5, letting Astra hold the global information, only feeding Fable partial information with session isolation, and keeping one Fable session that only gets global information for planning and isn't allowed to do any actual operations. I wonder if there's any chance this could get around it....

6
471

Comments (6)

Your avatar
Sign in to comment
Brielle
Brielle2 weeks ago

Don't rush to switch accounts or reinstall just yet—the criteria can vary by region and account type, so first check whether others in your area are reporting the same thing.

紫兰
紫兰2 weeks ago

Honestly, this isolation approach is not easy to pull off. Fable's detection tends to look at the overall conversation pattern and request characteristics—even if you only feed it partial information, as long as the call path and context structure still line up, the flag will most likely trigger anyway. On top of that, the planning session itself generates recognizable traffic, so simply splitting roles may not change the detection outcome.

永生不灭地天使
永生不灭地天使2 weeks ago

I'm not sure switching accounts and reinstalling will actually fix the problem. If it's tied to the device or usage patterns, switching accounts might just delay things. Could you tell us whether the flag reproduces consistently or only occasionally? That way we can figure out whether it's behavior-triggered or an environment issue.

国际语音短信
国际语音短信2 weeks ago

Wiping everything, reinstalling, and then reworking the architecture would cost a lot of time—and it might all be for nothing.

丽娜
丽娜2 weeks ago

In reality, many platforms' risk controls accumulate across sessions, so session isolation alone may not be effective. If you really want to try it, I'd suggest validating on a small scale first and noting which actions trigger the flag, rather than changing the entire environment right away. Also, official criteria for this kind of judgment are usually not made public, so it's best to watch for any updated documentation or more recent feedback in the community.

用户倾城之恋
用户倾城之恋2 weeks ago

Was it getting flagged even when you were only running planning, or did it only trigger during actual operations? That distinction is pretty important—it determines whether the isolation approach is on the right track.