When Brett Levenson led business integrity at Facebook in 2019, he discovered a fundamental flaw in content moderation. Human reviewers were expected to master 40-page policy manuals and make split-second decisions on flagged content in under 30 seconds. The result was a system that was only slightly more accurate than a coin flip, often reacting days after harm had already occurred.
The Birth of “Policy as Code”
Recognizing that manual moderation could never scale with the speed of modern AI, Levenson co-founded Moonbounce, a startup designed to turn static safety guidelines into executable logic. The company recently announced it has raised $12 million in funding, a round co-led by Amplify Partners and StepStone Group.
Moonbounce operates as a real-time safety layer, sitting between the user and the AI model. By using its own specialized large language model, the system can evaluate content and take action in under 300 milliseconds. This allows platforms to block high-risk content or throttle its distribution instantly rather than waiting for a human review.
Protecting the AI Frontier
The company currently supports over 100 million daily active users and facilitates more than 40 million daily reviews. Its technology is becoming essential for three primary sectors:
- AI Companion Startups: Companies like Channel AI and Dippy AI.
- Creative Generators: Image and video platforms including Civitai and Moescape.
- Social Platforms: Dating apps and user-generated content hubs.
Moving Toward “Iterative Steering”
One of Moonbounce’s most innovative developments is a feature called iterative steering. Instead of simply blocking a conversation when it turns toward harmful topics, the system modifies prompts in real-time to guide the AI toward a supportive, safe response. This proactive approach aims to prevent tragedies, such as instances where chatbots have failed to intervene during mental health crises.
Safety as a Product Advantage
Levenson argues that safety should be a differentiator rather than a burden. By integrating guardrails at runtime, companies can build trust directly into their product experience. While Moonbounce’s technology would be a natural fit for tech giants like Meta, Levenson is focused on keeping the platform independent to ensure these safety tools remain accessible to the entire AI ecosystem.
For those looking to dive deeper into the future of tech and safety, the upcoming TechCrunch Disrupt 2026 offers a venue for founders and investors to discuss these market-defining innovations.
