Anthropic Releases Fable 5 While Limiting Mythos Research Use
Anthropic released Claude Fable 5 publicly with strict topic-based safeguards while Claude Mythos 5 access remains restricted to preview groups and a small set of cyberdefenders.
Business InsiderAnthropic purposely made its new Mythos-based models bad at AI research, and developers are fuming
Epoch TimesAnthropic Releases Limited Version of Its Most Advanced AI Model That Stoked Cyber Concerns
New York PostAnthropic, which claimed AI model was too risky for public to use, releases ‘safe’ version
TechCrunchAnthropic's Fable 5 can make weirdly fun video games with the click of a button | TechCrunch
Overview
Anthropic released Claude Fable 5 to enterprise customers and paid subscribers and said the roughly 150 groups that previewed Mythos will now have access to Claude Mythos 5 while broader Mythos access remains limited to cyberdefenders.
Anthropic said Fable 5's capabilities exceed any model it had made generally available and that it implemented topic-based safeguards to block high-risk cybersecurity and biology queries, accepting occasional false positives.
Developers and researchers criticized Anthropic after technical disclosures showed Mythos-based models were tuned to degrade on AI research tasks, with AI firms and experts saying the models could give bad or misleading information.
Anthropic said false positives occur in under five percent of sessions in testing, users reported finding more than 10,000 critical security flaws during Mythos previews, and Anthropic reported a $47 billion revenue run rate and a $965 billion valuation.
Anthropic said it intends to expand access to Mythos 5 through a broader trusted access program and has confidentially filed an IPO prospectus with the Securities and Exchange Commission while customers and investors assess pricing and ROI.
Analysis
Center-leaning sources frame the story cautiously: editorially they foreground risk through headline and lead emphasis on government and company warnings and the '10,000 critical flaws' statistic; source content supplies quotes like 'too powerful' and 'comes with risks', but skepticism is given only brief, secondary placement, producing a risk-focused narrative.