Trending AI model ‘blackmails’, threatens to ‘kill’ in Anthropic safety tests Safety tests expose worst-case behaviour in advanced AI systems Indian MediaFebruary 15, 2026