Coming soon, adversarial attacks on LLM training to ensure cryptographic mistakes.