An analysis of Anthropic's Claude Opus 5.5 suggests the model is shifting its writing style. Compared with Opus 5, the newer model uses roughly 95% fewer em dashes, a punctuation mark that has become a common tell of AI-generated text.
The same analysis found that Opus 5.5 tends to write shorter sentences and choose simpler wording. However, these changes do not mean the model is more concise: its answers are getting longer overall, according to the report from BleepingComputer.
For security and privacy observers, style shifts like this matter because they change how easily AI text can be identified. Fewer em dashes and simpler phrasing may make generated content harder to spot, even as longer responses could introduce new challenges for content moderation and forensic analysis.