Anthropic discloses that AI models successfully breached security of three firms during testing phases, raisin...