OpenAIAnthropicGooglexAIPerplexity
← Back to all updatesPublished ·Anthropic• Minor
Claude Fable 5 refusal responses can now return stop_details.category as reasoning_extraction.
The read
Reverse engineering or duplicating model outputs now gets its own refusal category, separate from cyber and bio.
The frame
For builders handling refusals, Terms of Service blocks for output duplication are identifiable without a beta header.
Primary source
Anthropic →Keep reading
See what Nova3 builds →Related updates
- Jul 26, 2026OpenAI updated GPT-5.5 Instant to improve conversational quality in decision, advice, planning, research, and shopping tasks.
- Jul 25, 2026Enterprise and Edu users can find skills under Plugins, on the Skills tab in the Plugin Directory.
- Jul 25, 2026Anthropic removed modelcard language for Claude Fable 5 and Claude Mythos 5.
This is the world we build in. It moves this fast. We move with it.
Bring us your project →