Claude Opus 4.7 identifies authors from unpublished texts with minimal input
Updated
Updated · The Washington Post · Apr 26
Claude Opus 4.7 identifies authors from unpublished texts with minimal input
5 articles · Updated · The Washington Post · Apr 26
The AI model accurately named Kelsey Piper and Megan McArdle as authors using as few as 124 words from unpublished works.
These tests highlight the risk that AI can de-anonymize writers, threatening privacy for anyone with personal writing online.
Experts warn this capability could endanger anonymous sources, political dissidents, and vulnerable individuals seeking help, fundamentally challenging the future of online anonymity.
Can AI reliably identify authors, or will it lead to false accusations?
How can journalists protect sources when AI can identify them by their prose?
Is AI's ability to kill anonymity a bug or an intended feature?
If your writing style is a digital fingerprint, is online anonymity now impossible?
With AI de-anonymizing the web, what happens to online support groups?
Will AI 'synthetic dissidents' become the new face of online protest?
How Claude Opus 4.7’s 1 Million Token Context Window Revolutionizes Author Identification
Overview
In April 2026, Anthropic released Claude Opus 4.7, a groundbreaking AI model capable of identifying authors from small, unpublished text samples with remarkable accuracy. This ability, powered by advanced technical features like a massive 1 million token context window and enhanced vision, outperformed rivals such as ChatGPT and Gemini. The model’s skill in detecting unique writing styles sent shockwaves through communities relying on anonymity, raising serious privacy and ethical concerns for journalists, activists, and whistleblowers. Looking ahead, Anthropic’s upcoming Mythos model promises even greater capabilities, intensifying privacy risks and prompting the development of countermeasures and calls for new legal protections to safeguard anonymous expression.