k
keynews.ai
keynews.ai
WorkThe story, in brief

Claude Opus 4.7 has turned into an overzealous query cop, devs complain

Claude Opus 4.7's refusal rate just spiked. Developers are paying for answers they're not getting.

Reporting by The Register AI/ML 1 min brief
Illustration of two anonymous hands arranging task cards around an amber tool on a shared desk.
People, judgement and the changing nature of work.AI illustration by KeyNews
The KeyNews take

Why it matters

Anthropic's safety classifier appears to be over-filtering legitimate queries in Claude Opus 4.7, creating customer friction and raising questions about the tradeoff between safety guardrails and usability at production scale.

The key facts

5 to know

  1. Claude Opus 4.7 experiencing elevated refusal rates

  2. Issue attributed to Acceptable Use Classifier behavior

  3. Developer complaints about paying for non-functional responses

  4. Safety-vs-usability tradeoff becoming visible in production

  5. Published Apr 23 2026 — recent industry feedback

Go to the source

The Register AI/MLtheregister.com

Publisher excerpt: Rising refusal rate from Acceptable Use Classifier leaves customers paying for nothing
Read original report
Back to today's editionMore work news

Keep reading

Related stories

More from Work
Work

Following: The government doesn't want to regulate AI

AI lab leaders are calling for regulation. Washington isn't listening.

PlatformerWork

AI is exciting audit firms — maybe too much

KPMG and EY are rolling out AI audit tools faster than their risk controls can keep up. Here's what happens when the hype outpaces the guardrails.

Financial Times TechnologyWork

‘Flock City PD:’ The Fake Flock-Owned ‘Police Department’ That Searched Real Cameras for Real People

Flock Security's own demo exposed how its camera network could enable discriminatory surveillance—searches for religious symbols, political stickers, and vehicle types that mirror real-world bias.

404 Media