Skip to main content
Using Claude

Claude Switched Models Mid-Conversation? Understanding and Disabling Flagged Auto-Switching

Why Opus 5 and Fable 5 / 5.1 show "your message was flagged" and switch models — it's a safety mechanism, not an error. You can turn it off in Settings > Capabilities; with it off, flagged requests pause instead of switching.

ByUpdated
🌐 This article was machine-translated and may contain inaccuracies. Read the Korean original if in doubt.

While chatting with Claude Opus 5 or Claude Fable 5 / 5.1, you may see a notice that your message was flagged and the conversation has switched to a different model. This isn't an error or a billing issue — it's a safety mechanism working as designed on the top-tier models, and the automatic switch can be turned off in settings. Here's what happens, why, and what turning it off changes.

🟢 Current lineup (September 2026): Claude Opus 5.5 / Claude Sonnet 5 / Claude Haiku 4.5 (higher tier: Claude Fable 5.1). · model notice · Fable subscription
🔵 Current lineup (September 2026): Claude Opus 5.5 / Claude Sonnet 5 / Claude Haiku 4.5 (higher tier: Claude Fable 5.1). Parts of this article may still describe an earlier model.

Fable 5 and 5.1 subscription (updated September 7, 2026): Claude Fable 5.1, released September 1, 2026, is the current Fable model and Fable 5 is now legacy. Plan terms are the same for both — Max and Team Premium plans include Fable at up to 50% of the weekly usage limit; Pro and Team Standard use usage credits (

🔵 Current lineup (September 2026): Claude Opus 5.5 / Claude Sonnet 5 / Claude Haiku 4.5 (higher tier: Claude Fable 5.1). Parts of this article may still describe an earlier model.
0/M input, $50/M output tokens). The one-time
🔵 Current lineup (September 2026): Claude Opus 5.5 / Claude Sonnet 5 / Claude Haiku 4.5 (higher tier: Claude Fable 5.1). Parts of this article may still describe an earlier model.
00 credit applied only to the Fable 5 transition and is not offered for 5.1. Some coding and debugging requests may be answered by an Opus model due to a security classifier (both models). See the Fable 5.1 guide and the Fable 5 availability guide for details.

What happens when a message is flagged Your request (Opus 5 · Fable 5 / 5.1) Automatic safety check (runs on every request) Passes → answered as-is most requests Flagged → switches down Opus 5→4.8 · Fable 5 / 5.1→Opus Switching off? → the conversation pauses instead You choose to edit and retry (toggle in Settings > Capabilities)

Quick glossary

  • Flag — the automatic safety check (classifier) Claude runs on every request marking it as potentially risky.
  • Automatic model switching — instead of ending a flagged request with a refusal, a one-step-down model takes over and answers.

Why the switch happens

Claude Opus 5 runs automated safety checks on every request, and when a narrow set of higher-risk cybersecurity requests gets flagged, the conversation visibly falls back to Claude Opus 4.8, the next-most-capable model. Per the official documentation, most everyday security work passes without intervention. Claude Fable 5 / 5.1 adds a biology-related high-risk category, which falls back to Opus; if a request is blocked before any output starts, the conversation switches immediately.

In short, it's a trade-off — rather than cutting the conversation with a refusal, a less sensitive model carries on, so you can keep using the top model while misuse risk stays contained.

How to turn it off

Automatic switching is enabled by default the first time you select Opus 5 (or Fable 5 / 5.1). To turn it off:

  • Claude app: Settings > Capabilities > toggle "Switch models when a message is flagged" off.
  • Claude Code: Config > MODEL & OUTPUT, same toggle.

With it off, a flagged request pauses the conversation instead of switching, and you choose whether to edit the message and retry. If knowing exactly which model answered matters for your work, turn it off; if uninterrupted flow matters more, leave it on.

If legitimate work keeps getting flagged

If a request seems unrelated to cybersecurity but gets blocked, or your legitimate security work keeps falling back, report it with Send feedback — false-positive reports feed into narrowing these safeguards. For defensive security use cases that are repeatedly affected, there is an application path to the Cyber Verification Program (CVP) for Opus.

Don't confuse it with usage-limit switching

On Pro and Max plans, hitting the Opus usage cap and dropping to a lower model is a separate mechanism (a plan/usage matter). This article covers switches triggered by safety checks regardless of usage; for limits, see the usage limit reset guide. Note that model switching works differently on the API.

Sources

Was this helpful?

Keep reading

Is Claude's Korean Really Getting Worse? What's Actually Going On

Some users say Claude's Korean answers feel off lately. Here's what's officially documented versus what's likely a language-switching bug, context compaction, or prompting issue — plus fixes to try now.

40

Anthropic's Claude Misuse Reports: The Major Cases Explained

Anthropic's threat intelligence reports document real cases of Claude being misused for hacking, fraud, and influence operations — and how Anthropic detected and stopped them.

46

Fixing the Claude Payment Error "This Card Type Is Not Allowed"

Here's why Claude payments show "this card type is not allowed," which cards actually work according to official support docs, and how to get a Korean card through checkout.

119

Do 'One-Shot 100%' Claude Jailbreak Prompts Actually Work?

Prompts claiming to bypass Claude's safeguards in one shot keep circulating online. Here's why they can't reliably work, and what actually helps when Claude declines a request.

130