Anthropic PBC is limiting the release of its latest artificial intelligence model to a handful of major technology firms, warning that the system may be capable of powering cyberattacks if software makers don’t have a chance to test it against their own defenses first.