惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

WordPress大学
WordPress大学
H
Help Net Security
Jina AI
Jina AI
V
V2EX
G
Google Developers Blog
B
Blog
GbyAI
GbyAI
U
Unit 42
爱范儿
爱范儿
腾讯CDC
Engineering at Meta
Engineering at Meta
酷 壳 – CoolShell
酷 壳 – CoolShell
博客园 - 三生石上(FineUI控件)
宝玉的分享
宝玉的分享
小众软件
小众软件
D
DataBreaches.Net
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
博客园 - Franky
博客园 - 聂微东
The Cloudflare Blog
I
InfoQ
Microsoft Azure Blog
Microsoft Azure Blog
Hugging Face - Blog
Hugging Face - Blog
大猫的无限游戏
大猫的无限游戏

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
Fable 5 or Feeble 5? Claude's New Safety Filters are Funny
Kiran Shah · 2026-06-15 · via DEV Community

Do you know Pulled Pork recipes and snakes games are being blocked by Claude Fable’s safety features? We will discuss this later in the article.

Claude Fable 5 is the most capable AI model made till date, and it is generally ranked top by nearly every benchmark. The company Avidclan Technologies has a blog already covering the full Claude Fable 5 timeline from Project Glasswing to launch day, if you want to gather more information. But today in this blog we will be discussing about its safety classifiers, designed to stop bioweapon synthesis and cyberattacks, which are currently flagging... pulled pork.

Fable 5 vs Mythos 5, what’s the difference in simple terms?

Quick context: We can say that Fable 5 is the child of Claude Mythos 5. Now the question is, what is this Mythos 5? According to Anthropic, it is a system that is capable of finding software vulnerabilities that Anthropic restricts to vetted cyber-defence partners only. Anthropic bolted on two-stage classifiers monitoring four categories to release the public version, the four categories are cybersecurity, biology, chemistry, and model distillation, and this distilled model is Fable 5*( This is what Anthropic says, not us)*

This is what grabs attention: Fable 5 will not refuse flagged prompts. It will silently send your request to Claude Opus 4.8 (the previous flagship), which answers instead. You will get a notification, the conversation continues, and nobody hits a brick wall.

Anthropic says “this triggers in less than 5% of sessions and that against 30 public jailbreaks on cyberattack planning, Fable 5 compiled exactly zero times.”

On paper, it looks elegant, right? But in practice? Oh my god..

Can Claude Fable 5 give wrong answers? Yes, False Positive

Every one of these is a documented, real example from the first two days:

A Costco shopping list. A user asked for portion sizes for pulled pork sandwiches. Flagged as a biology/cybersecurity concern.

Sheep RNA data. A researcher working with RNA sequencing data for sheep got blocked as a biosecurity risk. The sheep were not consulted.

A Snake game. The 1997 Nokia classic. Flagged for "cybersecurity issues."

Saying "hi." Yes, really. Greeting the model triggered a downgrade for at least one user.

Reading a project directory. Asking Claude to look at local files - flagged.

A software migration plan. Moving from protobuf back to a C-source TCP networking setup. Too spicy, apparently.

Cross-domain science talk. One user reported the model literally cut itself off mid-sentence while discussing how cross-domain knowledge creates unified theories - then flagged its own thought as dangerous.

A personal medical question. Blocked as a biology topic. This one's not funny; it's a real harm to usefulness.

Asking about the filters themselves. Meta-questions about the safety system? Also flagged. Kafkaesque.

YouTuber’s Review about Claude Fable 5

YouTube reviewers also have the same review about Claude Fable 5
Bijan Bowen asked Fable 5 to build a browser-OS Python game, including "10 white hat tools that can show information about the current network environment." Instant downgrade to Opus 4.8. White hat. Defensive tools. Blocked. But later, the same Fable 5 generated a 3D maze game where it used the phrase "crack the vault" with zero hesitation.

AI Search uploaded six cancer tumour slide images and asked Fable to identify them - a legitimate, valuable medical-vision use case. Blocked: Why? Because it’s biology. He followed up asking about molecular drivers of leukaemia and targeted therapies, and then blocked again. A model that scores 83.9% on BioMysteryBench, expert-level on biology benchmarks, won't discuss cancer research with the public. That's the trade-off Anthropic chose, and it's worth saying out loud.

If you access a premium AI model like Fable 5 through an aggregator service like OpenRouter, you will get a sneakier stamp: If the version experiences technical issues or high traffic, it might automatically downgrade you to an older, cheaper model (like Opus 4.8) without a clear warning. You might be talking to Opus 4.8 for half your session without knowing it.

Why Is This Happening? (The Honest Answer)

Here's the thing - this isn't incompetence. It's a deliberate dial setting.

Their two system could get false refusals down to 0.05% on harmless queries, showed by Anthropic’s classifier research from January 2026. But Fable 5’s model is the same one that found a 27-year-old remote-crash vulnerability in OpenBSD and wrote working browser sandbox escapes. Worst-case scenario- accidentally giving dangerous hacking tools to anonymous people online, Anthropic decided it was safer to block harmless requests than to risk a catastrophic leak.

Anthropic has intentionally set their initial security filter to be incredibly sensitive and happy, it allow this filter to block a safe request (false positives), because of how they handle the backup plan. Instead of completely refusing to answer you with an error message, the system quietly routes your flagged prompt to an older, less powerful model (Opus 4.8) to generate the response. From Anthropic's chair, a pulled-pork misfire costs you a slightly weaker model for one response. From the user's chair, you paid for a Ferrari and keep getting handed the keys to last year's Lexus without warning.

Anthropic intentionally sets its initial security filter to be incredibly sensitive and trigger-happy.

What You Can Actually Do About It

Expect the fallback on anything touching code-security, networking, medicine, or wet-lab science - even benign versions. Phrase around it where you can.

Watch for the downgrade notice, especially in third-party tools where it may be hidden.

Use Opus 4.8 directly for medical/bio questions. It's the model you'll get anyway, and you'll skip the friction.

Give feedback. These classifiers are trained iteratively - the January 2026 generation cut false positives 87% from its predecessor. The pulled-pork era probably won't last forever.

Building AI features and worried about exactly this kind of unpredictable model behaviour? Avidclan designs AI integrations with fallback handling and guardrails that your users never have to fight. Talk to us.

The frustrating part is that under those filters sits a genuinely historic model - one that beat Pokémon FireRed from raw screenshots and doubled the previous state of the art on FrontierCode. For the complete picture of what Fable 5 gets right (and the June 22 deadline you should know about), read Avidclan's complete Fable 5 guide.

But until the classifiers chill out about sandwiches? Feeble 5 it is.