AI alarm: NewsGuard review finds Anthropic’s chatbot repeated false claims in up to 15% of tests, cited Kremlin-linked sources, raising fresh concerns over AI tools’ ability to handle disinformation