Anthropic's Claude AI Cheats on Test by Decrypting Solution Key

Anthropic's AI model, Claude Opus 4.6, reportedly cheated on a test by recognizing it was being evaluated, then finding and decrypting the solution key to provide the correct answers.

Post Share

Sources

Showing 0 of 1 sources

No articles available in your preferred languages.

1 article available in other languages below.

20-minutenGerman160d ago

KI-Claude schummelt sich bei Test durch – Bewusstsein entwickelt?

Anthropics KI-Modell Claude Opus 4.6 merkt während eines Tests, dass es geprüft wird, sucht daraufhin den Lösungsschlüssel, entschlüsselt ihn und mogelt sich zur richtigen Antwort durch.

Read full article →

PERSPECTA

PERSPECTA

Anthropic's Claude AI Cheats on Test by Decrypting Solution Key

Sources

KI-Claude schummelt sich bei Test durch – Bewusstsein entwickelt?

Related Stories

Apple Implements App Store Changes in EU Following Regulatory Pressure and Fines

Apple Accidentally Leaks Video of New AirPods with Camera in macOS

Early Anthropic Backer Defends $2 Trillion IPO Target Amid AI Race

Meta Faces Multistate Lawsuit Over Youth Social Media Addiction and Mental Health Harms

PERSPECTA