Anthropic's Claude Fable 5 returned to general availability after a Commerce Department ban lift, but users report widespread performance issues stemming from overly aggressive safety guardrails that frequently route tasks to the less capable Opus 4.8 model.
Anthropic's Claude Fable 5, the company's most powerful AI model, returned to general availability this week following a ban lift by the Department of Commerce — but early user reactions have been sharply critical, with widespread complaints about degraded performance and overly aggressive safety restrictions.
The model is now accessible to subscribers on Max, Pro, and Team plans, though Anthropic has imposed significant usage caps. Under the current structure, Fable 5 is limited to 50% of a user's weekly usage allotment. Beginning July 7, the model will transition to a pay-to-play system requiring usage-based credits, further restricting access for flat-rate subscribers.
Reports surfacing on Reddit describe a model that feels substantially weaker than its original release, with users pointing to frequent fallbacks to Anthropic's Opus 4.8 model when Fable 5 encounters prompts it deems sensitive.
"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. "This is not the model that got banned."
The degradation appears to affect both the Claude desktop application and Claude Code, Anthropic's developer-facing coding tool. One user reported that Fable "didn't even let me search for dead code without switching to Opus," while another said it was "very very obvious" when the fallback triggers because the system announces the shift visibly to the user.
A developer working on systems-level programming said the model was effectively unusable for certain tasks, noting that references to languages including C, C++, Rust, and Win32 API — as well as files containing words such as "security," "vulnerable," "unsafe," or "hook" — appeared to trigger a fallback or outright block.
Anthropic has not publicly acknowledged the reports of false positives. However, the company has noted that its updated safeguards operate with a large "safety margin," which may account for the elevated rate of spurious fallbacks users are experiencing.
Contrary to some user interpretations, the model itself does not appear to have been structurally altered. The more likely explanation, based on available information, is that Anthropic applied tighter safety guardrails to the restored release — guardrails that are triggering more broadly than intended, including on tasks that present no apparent safety risk.
The relaunch comes amid a broader policy backdrop. The Trump administration has imposed restrictions on private AI model releases, a move that has accelerated interest in open-source alternatives across the industry. Fable 5's return to availability following the Commerce Department's ban lift illustrates the regulatory complexity now surrounding frontier AI deployments.
Anthropic has not set a public timeline for addressing the false-positive issue, though the company is widely expected to revise the guardrail calibration in a forthcoming update. Whether the model can recover its prior reputation for high-capability performance — particularly among developers doing security-adjacent or systems-level work — will depend on how quickly those adjustments arrive.
TechEchelon Staff bylines are produced collectively by the newsroom for short, breaking, and wire-style coverage. Longer-form reporting is published under the responsible reporter's name.
More from the Staff →Microsoft launched Microsoft Frontier Co. on Thursday, committing $2.5 billion and 6,000 employees to a new unit focused on hands-on artificial intelligence deployments for enterprise clients, joining Amazon, OpenAI, and Anthropic in the forward deployed engineering push.
Meta Platforms shares rose 8% after the company confirmed plans to build a cloud business selling its surplus AI computing capacity to outside customers, a move that sent rival neocloud stocks sharply lower.
Tesla is testing a production Cybercab without a steering wheel or pedals in Austin, Texas, with a safety monitor aboard — a step toward the company's broader robotaxi ambitions as a key federal regulatory change nears approval.
Written for readers who already know the basics — markets, AI, and the policy decisions that shape both.