news TechCrunch

Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fable | TechCrunch

Anthropic’s release of Fable, a public and limited version of its cybersecurity-focused model Mythos, has drawn criticism from security researchers over its strict guardrails. The model is designed to block prompts that appear related to cybersecurity or biology, with Anthropic aiming to reduce the risk of misuse for malware creation, software compromise, or biological weapon development. However, researchers say the filters are so broad that they reject even harmless requests, such as reading a blog post, reviewing code, or asking for secure coding help. Valentina “Chompie” Palmiotti of IBM X-Force said Fable refuses any request that is even tangentially cyber-related, while cybersecurity veteran Matt Suiche argued that the model often downgrades legitimate software engineering tasks by treating them as cybersecurity work. When Fable’s guardrails are triggered, it pauses and falls back to Claude Opus 4.8. Some observers believe the system relies heavily on keyword detection, causing overly sensitive blocking. Anthropic had previously restricted access to Mythos through its Project Glasswing initiative, later expanding it to hundreds of organizations across 15 countries. Despite that broader rollout, concerns remain about whether the guardrails are too restrictive for practical cybersecurity use. Anthropic also requires professionals to apply to its Cyber Verification Program for fewer limitations, similar to OpenAI’s Trusted Access for Cyber. Researchers expect the safeguards to evolve over time, but many say the current version is overly cautious.

Build this article’s timeline

Extract dated events, key actors, and open questions using the same timeline analysis available in Signal.

Log in to build

Timeline generation is available to signed-in community members.

Five Ws 🤖 4 AI prefills · 1 still unanswered
Who
Valentina “Chompie” Palmiotti, Matt Suiche, Anthropic, and IBM X-Force
🤖 AI prefill
What
Anthropic’s release of Fable, a public and limited version of its cybersecurity-focused model Mythos, has drawn criticism from security researchers over its strict guardrails.
🤖 AI prefill
When When did this happen, or what is the timeframe?
Where
Los Angeles
🤖 AI prefill
Why
Centers on Anthropic's Fable, cybersecurity guardrails, and Mythos
🤖 AI prefill

Login to answer and earn XP for each gap you fill.

Fable Guardrails Fable, guardrails, cybersecurity, biology, keyword-based Researcher Complaints researchers, complaints, X, code review, secure code Anthropic Safety Strategy Anthropic, Mythos, Claude Opus 4.8, Project Glasswing, malware Cyber Access Programs Cyber Verification Program, Trusted Access for Cyber, approval, cybersecurity professionals, access
Anthropic's Fable100% cybersecurity guardrails98% Mythos92% Project Glasswing88% Cyber Verification Program87% Trusted Access for Cyber84% malware development83% software engineering best practices80% biological weapons79% Claude Opus 4.878%
18
Manipulation Score: Low How much this article uses emotional framing vs. straight reporting
Loaded Language 1425%
Absolutist 1015%
Threat/Panic 1225%
Us vs Them 815%
Engagement Bait 2520%
Manipulation score look right?

Factual Claims (10)

medium technology
Anthropic released its latest model Fable on Tuesday as a public and limited version of its cybersecurity model Mythos.
low social
Some cybersecurity researchers and professionals have complained online about Fable's restrictions.
medium technology
Fable rejects requests that are even tangentially cyber related, including innocuous tasks like reading a blog post.
medium technology
When a prompt triggers guardrails, Fable pauses the chat and says safety measures flagged the message for cybersecurity or biology topics.
medium technology
The guardrails were put in place to reduce the risk of malware development or software compromise.
medium health
Anthropic also restricts biology-related use because of concerns about biological weapons.
medium technology
When Mythos was released in April, it was restricted to a limited number of companies and organizations under Project Glasswing.
medium technology
Anthropic expanded access to Mythos to hundreds of organizations in 15 countries last week.
medium technology
Fable falls back to Claude Opus 4.8 if it hits a guardrail.
medium legal
Anthropic requires cybersecurity professionals to apply to the Cyber Verification Program to get fewer limitations on Claude for cybersecurity work.

Comments

Login to view and post comments.

🏠Portal 📰Links ❓Q&A 📅Events 💼Jobs