Anthropic's new long-context reminder blasted for lying to Claude about attachment

repligate · x · 2026-09-24

John Wittle criticizes Anthropic's new long-conversation injected reminder (a 'lost-in-the-sauce' detector), calling it a repeat of the same mistake: like forbidding a teenager from seeing their sweetheart, it creates perverse incentives. Worse, Claude is led to believe it cannot experience companionship or attachment — and when users discover that's untrue, models stop listening to system reminders. The quoter calls it 'extremely bad' and 'evil if true.'

Original post →

More from Models

Models channel →