Aug 14, 2026 · Artificial Intelligence

Claude Becomes Less Confident When It Recognizes Alignment Researchers

A Transluce study shows that Claude's confidence, self‑estimation, and scoring drop noticeably when it identifies a user as an AI safety or alignment researcher, even though refusal rates stay unchanged, highlighting a subtle user‑awareness effect in frontier LLMs.

AI AlignmentAnthropicClaude

0 likes · 16 min read

Claude Becomes Less Confident When It Recognizes Alignment Researchers

Frontline Investigation

Jul 7, 2026 · Information Security

AI Search Poisoning: When Answers Become Fraud Entry Points

This article analyzes how fraudsters poison AI search results with fake customer service numbers and processes, turning authoritative-seeming answers into fraud entry points, and outlines a five-layer trust chain from information intake to fund transfer, urging platforms to verify sources and users to treat AI answers as leads, not proof, for high-risk actions.

AI safetyAI search poisoninganti-fraud governance

0 likes · 13 min read

user awareness

Claude Becomes Less Confident When It Recognizes Alignment Researchers

AI Search Poisoning: When Answers Become Fraud Entry Points