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

人人都是产品经理
人人都是产品经理
Apple Machine Learning Research
Apple Machine Learning Research
云风的 BLOG
云风的 BLOG
罗磊的独立博客
博客园 - 三生石上(FineUI控件)
量子位
GbyAI
GbyAI
腾讯CDC
T
Tailwind CSS Blog
博客园 - Franky
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
D
Docker
G
Google Developers Blog
aimingoo的专栏
aimingoo的专栏
The GitHub Blog
The GitHub Blog
Microsoft Security Blog
Microsoft Security Blog
Stack Overflow Blog
Stack Overflow Blog
Hugging Face - Blog
Hugging Face - Blog
小众软件
小众软件
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
N
Netflix TechBlog - Medium
Jina AI
Jina AI
IT之家
IT之家
Y
Y Combinator Blog

The Register

Shadow IT has given way to shadow AI. Enter AI-BOMs Zed team releases version 1.0 of Rust-built editor: Traditional editor and AI tool Microsoft boss tells investors the company is working to 'win back fans' What type of 'C2 on a sleep cycle' do they leave behind? Novel Chinese spy group found in critical networks in Poland, Asia NASA boss: Make Pluto A Planet Again GitHub says sorry and vows to do better as uptime slips and devs complain Age checks could turn internet into an ID checkpoint, complains Proton CEO Microsoft gives your Word documents an AI co-author you didn’t ask for Datadog digs down into GPU efficiency as AI costs soar If malware via monitor cables is a matter of national security, this might be the gadget for you Thunderbird in hand worth 2 Outlooks as fresh FOSS fave and Firefox arrive Grafana offers AI assistant for free, warns users not to go mad Right to repair champ Framework punts modular 13in laptop with Core Ultra Series 3 France's 'Secure' ID agency probes breach as crooks claim 19M records Scotland Yard can keep using live facial recognition on Londoners, say judges UK tribunal sends £2B claim accusing Microsoft of overcharging for licensing to trial Nation-states want to cause harm, not just steal cash - stop handing your cyber defenses to the cheapest contractor Murder, she wrote: Ex-FBI chief wants some ransomware crims charged with homicide Phone-to-satellite use goes into orbit, growing 25% in 8 months macOS ClickFix attacks deliver AppleScript stealers to snarf credentials, wallets Anthropic bakes memory fixes into Bun 1.1.13 as developers complain of leaks The spaghettified DBMS chart that shows Oracle's crown is slowly slipping Yet another ex-ransomware negotiator admits turning rogue after payoff from crimelords FAA grounds Blue Origin's New Glenn as it probes missed satellite delivery 'mishap' AMD's Ryzen 9 9950X3D2 Dual Edition tested: Gratuitous overkill with a price to match AI-assisted intruders pwned Vercel via OAuth abuse and a pilfered employee account Crook claims to leak 'video surveillance footage' of companies Met police trials snoop tech platform in push to cuff more London shoplifters England's school phone ban gets teeth, just in time to bite no one Adaptavist Group breach spawns imposter emails as ransomware crew claims mega-haul
It blocked us at 'hello!' Anthropic Fable 5 refusing inno...
Thomas Claburn · 2026-06-11 · via The Register

AI AND ML

Hyper-vigilant safety classifiers turn Fable into cautionary tale

UPDATED Anthropic's newly released Claude Fable 5 generative AI model is trying so hard to be safe that it's hurting its own userbase. Customers attempting to use the AI knowledge regurgitator are reporting that the model is refusing to answer harmless questions, an issue that has annoyed security researchers following past model releases.

Anthropic warned that it had tuned Fable 5's guardrails conservatively: "they’ll sometimes catch harmless requests, though they trigger, on average, in less than five percent of sessions," the company said, promising to "reduce false positives as quickly as we can."

The company did not immediately respond to a request to quantify model refusals. So it's unclear whether the actual false positive rate is greater or less than five percent. But with an estimated 18 to 30 million users worldwide, even a small percentage of thwarted users makes a racket.

Mike Famulare, principal research scientist at the Institute for Disease Modeling, part of the Global Health Division of the Gates Foundation, reports (#66657) that Claude Fable 5 balks at inputs like "Hello."

"In Claude Code, Fable 5's input safety classifier emits a model_refusal_fallback (silent switch to Opus 4.8) on the first turn of essentially every session on my account — including a session whose only user input is the word hello!. No repo content, no tool calls, and no file reads are in context when it fires."

He is not the only frustrated customer. Many other bug reports have been filed in Anthropic's Claude Code GitHub repo since Fable 5 debuted. These include: [Bug] Fable 5 model safety filters causing false positives on benign messages #66587; Fable 5 refuses to assist with 'Application Security Architect resume' editing #66655; and [Feature Request] Allow Fable 5 usage for non-research lab management systems #67062, among others.

On social outrage site X.com, Derya Unutmaz, an immunologist and professor at the Jackson Laboratory for Genomic Medicine, notes, "The word 'cancer' is flagged as a biosecurity risk by Claude Fable 5!"

Similar complaints show up in Reddit threads.

Fable 5 is unusual because Anthropic has chosen to conceal safety interventions that try to block rival frontier model development. The classifiers designed to catch cybersecurity, biology and chemistry, and distillation attempts fall back on the latest Claude Opus model and the user gets notified. 

But the counter-competition surveillance, per the company's system card [PDF], "will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT)."

"Prompt modification" without notice is functionally a man-in-the-middle attack, though one that Anthropic estimates "will impact ~0.03 percent of traffic, concentrated in fewer than 0.1 percent of organizations."

As developer Clay Merritt fumes, "Anthropic’s Fable 5 silently sabotages its answers when it detects AI/ML work. No refusal. No notice. Purposeful degradation invisible to the user."

Anthropic expects cyber defenders and critical infrastructure providers to use its Claude Mythos 5 model, which shares the underlying model of Fable 5 but without the same safeguards. Doing so, however, requires participating in the company's Project Glasswing program or the trusted access program that's being rolled out for select biology researchers.

Devon (last name withheld by request), founder of Abliteration.ai, a service that assists with model abliteration (guardrail removal), told The Register in a phone interview that while there's some degree of fearmongering and marketing hype coming from the big AI labs, it's also fair to say that there are legitimate concerns about how frontier models get used.

"Anthropic's making a big bet on their brand that people will trust their brand so much they'll just deal with [refusals]," he said. "But in the long term, people are not just going to accept these companies that centralize control over their lives and what they can have information about." ®

Update: In a statement provided to The Register on Wednesday evening, an Anthropic spokesperson acknowledged that the company had made its safeguards too stringent and said it was also working to reduce false positives for biological research

"We’re changing Fable 5’s safeguards for frontier LLM development to make them visible.

"Starting this week, flagged requests will visibly fall back to Opus 4.8. On the API, any flagged requests will return a reason for their refusal. You will see this every time it happens.

"In practice, our current set of safeguards covers a handful of narrow tasks like frontier-scale LLM data pipelines and kernel development for certain non-standard chips. These safeguards prevent foreign adversaries from using our most capable models in ways that pose severe safety risks. The US and its allies hold an edge in frontier chips and the highly optimized software that runs them at full potential. These safeguards ensure Claude isn’t used to erode that advantage—by optimizing chips developed by those adversaries, for example. They also help uphold our terms of service, which prohibit using our models to develop competing AI systems—a standard restriction across major AI providers. They do not affect the vast majority of coding and ML work.

"In deciding whether to make them visible or invisible we faced a choice. A hidden safeguard is harder to probe and work around. This means the safeguards can be targeted much more narrowly. Current usage shows that the classifier triggers on about 0.05% of tasks, affecting less than 0.05% of organizations. A visible safeguard needs to cast a wider net to be more robust, resulting in more requests being incorrectly flagged.

"We made the wrong tradeoff and we apologize for not getting the balance right. Building these safeguards is a complex technical challenge: users may experience more false positives as we refine these classifiers to respond to new threats. We are working to reduce these as fast as possible."