"The new guardrails are kicking in on way too many tasks and falling back to Opus 4.8," one user wrote in a Reddit post.
Anthropic's relaunch: availability, a $100 Max surprise, and usage caps
When the Department of Commerce announced it was lifting the ban on Claude Fable, Anthropic made Fable 5 available to all users — including those on the $100 Max subscription — but not without limits. According to Anthropic, Fable 5 is included in Max, Pro, and Team plans yet is "heavily capped." Users can run Fable for up to 50% of their weekly usage limits under current rules, and that access will change again after July 7, when the model will transition entirely to a pay-to-play arrangement using usage credits.
Safety guardrails, routing behavior, and Opus 4.8
Users report the restored Fable often routes tasks to Opus 4.8 or refuses requests altogether, producing what the community calls a "nerfed" experience. BleepingComputer reports that the model itself has not been nerfed; instead, Anthropic appears to be routing requests through stricter safety systems. The company has said its updated safeguards rely on a large "safety margin," a choice that could explain frequent fallback behavior. BleepingComputer also observed Fable being routed to Opus 4.8 even on tasks that did not appear to pose a safety risk.
Developer accounts: systems code, dead-code searches, and trigger words
Multiple developers on Reddit described practical failures in coding workflows. One user said Fable "didn't even let me search for dead code without switching to Opus," and another said it was "very very obvious" when a fallback triggered because Claude notifies the user and visibly shifts to Opus. A developer working on systems-level projects reported the model was effectively unusable for C, C++, Rust, Win32 API references and memory-related work; they said files containing words such as "security," "vulnerable," "unsafe," or "hook" appeared to trigger a fallback or block. Those accounts suggest the safeguards are highly sensitive to security-adjacent language and certain technical contexts.
Impact on day-to-day users and subscription holders
For people who were waiting to try the originally banned model, the relaunch delivered mixed impressions. Some users had prepared by topping up usage-based credit wallets, anticipating a strictly usage-based debut, only to find Fable available across tiers but limited to 50% of weekly usage and facing a July 7 shift to a credit-only model. For a model billed as the company's most powerful, the combination of caps and stricter guardrails has diminished practical access for many users.
What this means for developers, enterprises, and Anthropic
- Developers and systems programmers: Expect to switch between Fable and Opus for tasks that trigger guardrails; the reports indicate work on low-level languages and memory-sensitive code is particularly affected.
- Enterprises and procurement leaders: The 50% weekly cap and the upcoming jump to usage credits after July 7 will factor into budgeting and procurement decisions for teams that had planned to rely on Fable 5 for advanced tasks.
- Anthropic product and safety teams: Anthropic has not publicly acknowledged reports of false positives, but BleepingComputer says the company is likely aware and that an update to the safeguards may follow — a clear sign that product tuning remains underway.
The relaunched Fable 5 presents a clear trade-off: Anthropic's enlarged safety margin reduces perceived risk but also reduces practical usefulness for certain users and tasks. Users and teams who need consistent systems-level coding assistance are already reporting workarounds — switching to Opus, segmenting jobs, or avoiding certain language in prompts — while Anthropic's stated reliance on a "safety margin" raises a single, concrete question for the near term: will the company tune those guardrails to reduce false positives without sacrificing the safety aims that prompted the relaunch conditions in the first place? Anthropic has not yet publicly confirmed next steps, but BleepingComputer reports it is likely the company will address the issue in a future update.
Read the original reporting at BleepingComputer.