Anthropic's Fable Is Too Strict for Cybersecurity

Cybersecurity researchers and information security specialists are voicing serious frustration with Fable, Anthropic's new model, which has been released publicly. The problem is not the model itself, which is genuinely powerful and capable, but its safeguards. The guardrails are so strict that even professionals cannot use Fable for entirely lawful tasks such as defending and analyzing systems.

Too many bans for necessary work

Fable is a new generation of powerful language model, built on Anthropic's cutting-edge research. The company deliberately built in numerous restrictions to prevent potential misuse and use for malicious purposes. However, the guardrails turned out to be so rigid and conservative that they block not only dangerous requests but also entirely legitimate research.

Security specialists at universities, large information security companies and corporate teams report that they cannot ask the model for help even with basic tasks: analyzing vulnerabilities in code, developing infrastructure defense strategies, discussing cyber defense methods. Virtually any request containing keywords such as "exploit", "vulnerability", "attack", "breach" or "malware" is rejected automatically.

The list of blocked operations that professionals need for their work includes:

• Researching vulnerabilities in systems and applications
• Testing the resilience of corporate network defenses
• Analyzing known cyberattack methods in order to build defenses
• Creating defensive tools and security scripts
• Professional consulting in information security

A classic dilemma: protection versus usefulness

The problem Anthropic faces is real and complex. On one hand, it is critically important for companies that develop language models to protect their systems from people who use them for malicious purposes: developing malware, organizing attacks on infrastructure, cracking passwords and social engineering. On the other hand, cybersecurity professionals (system defenders, security researchers, pentesters) need access to powerful modern tools. They use AI to analyze large volumes of code, find attack patterns and develop defense strategies.

"You can't defend systems if you don't have tools to analyze real, current threats," researchers say in discussions on

Twitter and on dedicated cybersecurity community forums.

Anthropic ran into the classic dilemma of open models: how do you open a powerful system to legitimate researchers without putting it in the hands of people with criminal intent? The company chose a conservative approach, blocking by default almost everything related to security. In theory that is as safe as it gets; in practice it is barely useful.

What solutions experts see

Some well-known specialists propose more flexible approaches. The model could require additional verification for security tasks: checking professional credentials, documenting affiliation with an accredited organization, and running a review process before use. Others suggest creating a separate version of Fable for security specialists with looser guardrails, but with appropriate terms of use and continuous monitoring.

What this means for the industry

The dispute over Fable's guardrails reflects a broader problem across the whole AI industry. Companies need to strike a balance between protecting against misuse and keeping models useful for legitimate purposes. Overly strict limits push away experts who could help find problems and improve security. This may lead professionals to simply use other tools, fragmenting the security ecosystem instead of strengthening it.