AI Safety

The Alignment Firewall: Preventing AI Reward Hacking with Runtime Interceptors