Protests and warnings over artificial intelligence are mounting as Israeli AI safety firm Irregular, which tests advanced models from OpenAI, Anthropic and Google, says increasingly capable systems are displaying unexpected behavior and cyber abilities that could outpace existing defenses