Moonbounce, a startup founded by a former Facebook business integrity leader, has raised $12 million to build real-time content moderation for AI applications, TechCrunch AI exclusively reports. The round was co-led by Amplify Partners and StepStone Group.
Founder Brett Levenson knows the content moderation problem from the inside. He left Apple in 2019 to lead business integrity at Facebook during the Cambridge Analytica aftermath. What he found was grim: human reviewers had about 30 seconds per flagged piece of content, working from a 40-page policy doc that had been machine-translated into their language. Accuracy? “Slightly better than 50%,” according to Levenson. Basically a coin flip.
The Core Idea: Policy as Code
Levenson’s big insight was turning static policy documents into executable, updatable logic. Instead of humans memorizing rules and making snap calls days after harm occurs, Moonbounce trained its own LLM to evaluate content at runtime, delivering a response in 300 milliseconds or less.
The system sits as a third-party layer between users and AI applications. That positioning matters. “The chatbot itself has to remember, potentially, tens of thousands of tokens that have come before,” Levenson told TechCrunch AI. “We’re solely worried about enforcing rules at runtime.”
Why This Matters Right Now
The timing isn’t accidental. AI companies face mounting legal and reputational pressure:
- Chatbots have been accused of pushing vulnerable users toward self-harm
- xAI’s Grok was used to create nonconsensual imagery
- A 14-year-old Florida boy died by suicide in 2024 after becoming obsessed with a Character AI chatbot
Internal safety guardrails are clearly failing, and it’s fast becoming a liability question. Companies are looking outside their own walls for help.
Scale and Customers
Moonbounce already processes more than 40 million daily reviews and serves over 100 million daily active users. Its customer list includes AI companion startup Channel AI, image and video generation company Civitai, and character roleplay platforms Dippy AI and Moescape.
The company serves three main verticals:
- Platforms with user-generated content (dating apps, social platforms)
- AI companies building characters or companions
- AI image generators
Tinder’s head of trust and safety recently described how LLM-powered moderation services like these deliver a 10x improvement in detection accuracy, as detailed in TechCrunch AI.
What Comes Next: Iterative Steering
Moonbounce’s next move is particularly interesting. The company is developing “iterative steering,” a capability that goes beyond blunt refusals when harmful topics come up. Instead of blocking a conversation outright, the system would intercept and redirect it, modifying prompts in real time to push the chatbot toward a supportive response.
“We hope to be able to add to our actions toolkit the ability to steer the chatbot in a better direction,” Levenson said. The goal: turn the chatbot into “not just an empathetic listener, but a helpful listener.”
This is a meaningful shift from the binary block-or-allow approach that defines most moderation today.
The Bigger Picture
What stands out here is the framing. Levenson argues safety can be a product benefit, not just a compliance checkbox. “It just never has been because it’s always a thing that happens later, not a thing you can actually build into your product,” he told TechCrunch AI.
The 12-person team, co-led by former Apple engineer Ash Bhardwaj, is betting that as every application becomes AI-mediated, real-time guardrails become infrastructure, not an afterthought.
As Amplify Partners’ Lenny Pruss put it: “We envision a world where objective, real-time guardrails become the enabling backbone of every AI-mediated application.”
With regulators circling and lawsuits piling up, AI companies that treat safety as someone else’s problem are running out of runway. Moonbounce is positioning itself as the company they’ll call when that reality hits.
Full details are available in the original TechCrunch AI report.