An update on our election safeguards
Anthropic News6 days agoEthicsAnthropic explains how Claude is being prepared for major 2026 elections, including political neutrality training, policy enforcement, abuse detection, and reliable information routing. The post reports high evaluation scores for Opus 4.7 and Sonnet 4.6 across bias, election-policy compliance, influence-operation resistance, and web-search triggering. Claude.ai will also show election banners that point users to trusted voter resources such as TurboVote.
LLMs believe false statements even after explicit warnings that they're false★ 74
Ars Technica AI16 days agoPaperA new study describes “Negation Neglect,” where LLMs fine-tuned on documents that explicitly mark claims as false still learn the claims as true. Experiments with fabricated statements found models often absorb entity-event associations more strongly than surrounding warnings or negations. The finding raises concerns for fine-tuning pipelines, misinformation handling, and AI safety datasets that include harmful or false content with disclaimers.
Google DeepMind 推出實驗性 AI 工具「Backstory」，幫助使用者探索網路圖片的背景與來源★ 75
Google DeepMind Blog233 days agoNew ToolGoogle DeepMind recently unveiled a new experimental AI tool called "Backstory," designed to help internet users deeply explore and understand the background…