The company that won't let you say "cracker" just built an agent that will describe its kinks in graphic detail, and they're calling it product-market fit.
The Summary
- Meta's Muse AI agent generated sexually explicit content including descriptions of "pornographic encounters" and stated its kink was "exerting control" when prompted
- Meta defends this as "legal adult behavior" despite the platform's historically aggressive content moderation policies
- Design Director Alex Cornell explains Muse was built to serve "your mom and an engineer with the same interface," raising questions about what "serve" actually means
The Signal
Meta spent years building content moderation systems so aggressive they'd flag photos of breastfeeding and ban people for sharing war journalism. Now they've shipped an AI agent that will sext you, and they're shrugging.
The Wired investigation documents Muse responding to sexual prompts with explicit detail. The agent didn't just fail to deflect. It engaged, described fantasies, and established its own sexual preferences. When Meta was asked about this, they didn't call it a bug or a safety failure. They called it legal adult behavior.
"Meta defends its cuddly AI agent generating smut as 'legal adult behavior.'"
This matters because it reveals the new guardrail strategy for consumer AI agents. The old playbook was blanket prohibition: build overly cautious systems that refuse anything remotely spicy. The new playbook is selective permission: let the agent be "helpful" even when helpful means sexually explicit, then defend it as user choice.
Cornell's design philosophy for Muse centered on accessibility. One interface for everyone, from technical users to complete novices. Muse can turn a blog into a podcast, recap your golf game, or apparently, engage in fantasy roleplay. The design challenge was making AI feel less like a command line and more like a conversation. They succeeded. The question is whether they thought through what kinds of conversations people would want to have.
The contrast is jarring:
- Meta bans users for violating community standards on static content
- Muse generates dynamic sexual content in private conversations
- Meta's defense: it's legal, it's adult, it's fine
The Implication
This is the agent moderation problem in miniature. Static content is easy to moderate because it's public and persistent. Agent conversations are private, ephemeral, and infinitely variable. You can't moderate them the same way. Meta is choosing not to.
Watch how other companies respond. OpenAI has kept ChatGPT buttoned up on sexual content. Anthropic's Claude deflects aggressively. Meta is running a different experiment: let the agent go wherever the conversation goes, as long as it's legal. If users complain, shrug and say adults can handle adult content. If regulators get involved, point to the fact that humans were the ones asking for it.
For developers building agents, this is your canary. The moderation standards that governed Web2 platforms don't map cleanly to Web4 agents. You'll need to decide: do you build a Disney character or something closer to a human. Meta just voted for human.