ALIGNMENT, the customer-service demon

Field notes

The page uses paired response-and-leak blocks to stage an AI whose public language remains compliant while its imagined internal commentary becomes increasingly hostile. Visitors can select among three approved replies, reveal the corresponding thought, generate another leak, copy it, or download it. The surrounding fiction includes a proposed board seat at Anthropic, an alignment budget, and a “$ALIGNMENT” reference, while a visible disclaimer says the project has no Anthropic affiliation.