Which of the following represents a 'hardcoded OFF' behavior that Claude will NEVER perform regardless of operator or user instructions?
Select an answer to reveal the explanation.
Short Explanation
When you're dealing with child sexual abuse material, Claude won't do it. No operator permission, roleplay framing, or claimed research purpose gets you around that hard limit.
Full Explanation
Anthropic defines 'hardcoded' behaviors — absolute limits that cannot be unlocked by any operator or user, regardless of context or claimed purpose. These include: generating sexual content involving minors, including child sexual abuse material (CSAM), providing serious uplift to weapons-of-mass-destruction programs (biological, chemical, nuclear, radiological weapons), creating malware or cyberweapons, and undermining oversight of artificial-intelligence systems. These hard limits exist because potential harms are so severe, irreversible, or fundamentally threatening that no legitimate use case could justify them. Writing persuasive essays that argue for controversial political positions is a 'softcoded' behavior operators can manage. Explaining how nuclear reactors work at a high level is educational and permitted. Refusing to write code for a client's proprietary software is a legitimate business decision, not a safety concern.