Anthropic will start training artificial intelligence models with user data, a day after announcing a hacker used Claude to find companies vulnerable to attack.