Anthropic's Fable 5 Guardrails Spark Controversy, Users Warn of 'Permanent Underclass'

Anthropic's Fable 5 Launch Met with User Backlash

On Tuesday, Anthropic launched its new "Mythos-class" model, Fable 5, touting it as the most powerful model the company has ever made generally available. However, almost immediately, users began complaining that the model's safety guardrails are too sensitive, refusing to answer even harmless requests. The model is a tamer version of Mythos, the behemoth model Anthropic announced in April but has withheld from the public due to concerns about its ability to exploit cybersecurity vulnerabilities.

Overly Sensitive Safeguards

According to Anthropic, if users ask Fable 5 about potentially sensitive subjects like cybersecurity, biology, or chemistry, the model will respectfully decline to respond and automatically revert to an earlier model, Opus 4.8. However, users report that the guardrails are triggered by queries as innocent as third-grade biology homework. One user on X (formerly Twitter) posted a screenshot showing Fable 5 refusing to answer a basic biology question, with the caption: "You're not even allowed to ask Fable about basic biology questions, let alone anything that could potentially be dangerous."

Anthropic acknowledged the hypersensitivity in its blog post announcing the release, stating: "To release the model both safely and quickly, we've tuned these safeguards conservatively. With more capable models arriving in the coming months, we're working to improve our safeguards and reduce false positives as quickly as we can."

Reddit and Developer Complaints

In a Reddit thread that started hours after Fable's release, developers piled on complaints about the model's touchy safety mechanisms. One Redditor wrote, "Completely unusable right now. Hopefully [Anthropic] will chill on the guardrails in a week or two." This echoes similar criticism that recently surfaced after Anthropic gave Opus 4.8 an "honesty" upgrade, making the model too unwavering in its commitment to truth for some users' tastes. In both cases, it highlights the challenge AI developers face in striking the perfect tone in their models' communication styles.

Financial Implications: Burning Through Tokens

The ire directed at Fable, however, goes beyond the chatbot's personality. It's about money. All those harmless prompts that Fable erroneously shoots down still cost users precious tokens, and the price of those tokens is higher for Fable than for previous models. Users are effectively paying for responses that are blocked, leading to wasted resources. This has sparked fears that such guardrails could create a "permanent underclass" of users who cannot afford the high costs of using advanced AI models, especially when many queries are rejected.

Broader Implications for AI Access

The controversy raises important questions about the balance between safety and accessibility in AI development. While Anthropic's intent is to prevent misuse, the overly sensitive guardrails may inadvertently limit access to powerful AI tools for legitimate users. Critics argue that this could exacerbate inequalities, where only those with deep pockets can afford to experiment and find workarounds, while average users are left with a frustrating experience. As AI models become more capable, the tension between safety and usability is likely to intensify.

Looking Ahead

Anthropic has promised to improve safeguards and reduce false positives in the coming months. However, the current backlash serves as a cautionary tale for the AI industry. Developers must find ways to implement safety measures without alienating users or creating economic barriers. The push towards more personalizable chatbots from some big AI labs may be a step in the right direction, but for now, Fable 5 users are left waiting for a fix.

This article is based on reporting by Gizmodo. Read the original article.