- Moonbounce has raised $12 million to build an AI control engine that converts content moderation policies into executable, enforceable code.
- Founder Brett Levenson previously led business integrity at Facebook, where he found human reviewers were only “slightly better than 50% accurate” at applying content policies.
- The startup’s “policy as code” approach replaces static policy documents with programmable logic that can be updated and deployed in real time.
- The platform targets both social media companies and enterprises deploying AI chatbots that need consistent safety guardrails.
What Happened
Moonbounce, a startup founded by former Facebook business integrity lead Brett Levenson, has raised $12 million to scale its AI control engine that converts content moderation policies into consistent, predictable AI behavior. The funding was reported by TechCrunch on April 3, 2026. Levenson developed the concept after spending years at Facebook, where he observed firsthand the systemic limitations of human-driven content moderation at scale.
Before Facebook, Levenson worked at Apple, which he left in 2019 to join the social media company during the Cambridge Analytica fallout. He initially believed he could fix Facebook’s moderation problems with better technology, but soon realized the challenge was more fundamental than any single tool could address.
Why It Matters
Content moderation has become an increasingly urgent challenge as AI chatbots enter mainstream consumer use. High-profile incidents involving chatbots providing teens with self-harm guidance and AI-generated imagery evading safety filters have demonstrated that existing approaches are inadequate for the speed and scale of AI-generated content. Levenson’s experience at Facebook gave him direct insight into the structural reasons traditional moderation fails.
“It was kind of like flipping a coin, whether the human reviewers could actually address policies correctly, and this was many days after the harm had already occurred anyway,” Levenson told TechCrunch. At Facebook, human reviewers were expected to memorize a 40-page policy document that had been machine-translated into their native language. They then had approximately 30 seconds per piece of flagged content to make multi-dimensional enforcement decisions about whether to block the content, ban the user, or limit distribution. That reactive, delayed approach is not viable against well-funded adversarial actors who can generate and distribute harmful content faster than humans can review it.
Technical Details
Moonbounce’s core innovation, which Levenson calls “policy as code,” treats content moderation rules as programmable logic rather than static documents that humans must memorize and interpret. This approach allows policies to be version-controlled, tested against known edge cases, and deployed in real time, similar to how software engineers manage application code through continuous integration and deployment pipelines. The system converts natural language policy statements into structured decision trees that can be executed at machine speed with consistent outcomes.
The platform is designed to handle multiple enforcement actions simultaneously, including blocking content, suspending users, and throttling distribution, rather than forcing a binary allow-or-deny decision. This granularity addresses a core limitation Levenson observed at Facebook, where reviewers achieved accuracy rates only “slightly better than 50%,” effectively making their enforcement judgments no more reliable than random chance across the full spectrum of policy violations.
Moonbounce targets two distinct markets: traditional social media content moderation, where platforms must police user-generated content at scale, and the newer challenge of governing AI chatbot behavior, where conversational safety requires real-time evaluation of dynamic, multi-turn interactions rather than review of static posts or images. The latter market has grown rapidly as enterprise companies deploy customer-facing AI assistants that must comply with industry-specific regulations and brand safety requirements.
Who’s Affected
Social media platforms, enterprise companies deploying customer-facing AI chatbots, and AI model providers all represent potential customers for Moonbounce’s technology. The startup directly competes with internal trust and safety teams at major tech companies, as well as other startups in the AI safety and moderation space. Human content moderators, a workforce that has faced widespread criticism for poor working conditions and documented psychological harm from reviewing disturbing material, could see their roles evolve as automated policy enforcement becomes more reliable and reduces the volume of content requiring human review.
What’s Next
Moonbounce plans to use the $12 million funding round to expand its engineering team and accelerate product development. The company is entering a market where regulatory pressure is mounting, with the EU AI Act and various U.S. state-level AI safety laws creating new compliance requirements that companies deploying AI systems must meet. Levenson’s “policy as code” framework positions Moonbounce to serve organizations that need to demonstrate auditable, consistent enforcement of safety policies to regulators, a requirement that is likely to become more stringent as AI systems become more widely deployed.
