Moonbounce Raises $12 Million to Revolutionize AI and Content Moderation with “Policy as Code” Approach
In 2019, Brett Levenson left Apple to join Facebook’s business integrity team, eager to tackle the social media giant’s mounting content moderation woes. Fresh out of Silicon Valley’s tech powerhouse, Levenson believed advanced technology could swiftly resolve the platform’s struggles. But what he encountered was far more complex: a system where human reviewers, armed with hastily translated policy documents, were forced to make high-stakes decisions in mere seconds—often with accuracy no better than a coin toss.
Fast forward to 2026, and Levenson is leading Moonbounce, a startup that has just secured $12 million in funding to transform how companies moderate content, particularly in the age of generative AI. Backed by Amplify Partners and StepStone Group, Moonbounce is pioneering a “policy as code” approach, turning static rules into executable logic that can enforce policies in real time. Its mission? To prevent harmful content from proliferating online—whether it’s generated by users or AI chatbots—before it causes damage.
The Content Moderation Crisis
When Levenson joined Facebook, the platform was reeling from the Cambridge Analytica scandal and grappling with systemic issues in content moderation. Reviewers were expected to navigate 40-page policy documents, often translated by machines, and adjudicate flagged content in 30 seconds or less. These snap decisions—whether to block, ban, or limit content—were only slightly better than random, Levenson revealed.
“It was like flipping a coin,” he told TechCrunch. “By the time reviewers acted, the harm had already been done.”
This reactive, delayed approach has proven unsustainable as online platforms face increasingly sophisticated adversarial actors. The rise of generative AI has only exacerbated the problem, with chatbots offering self-harm advice to teenagers and AI-generated imagery bypassing safety filters. Such incidents have led to public outcry, regulatory scrutiny, and lawsuits, forcing companies to rethink their moderation strategies.
Enter Moonbounce: Policy as Code
Moonbounce’s solution lies in its ability to operationalize policies as code, embedding enforcement directly into the content generation process. Leveraging its proprietary large language model (LLM), the company evaluates content in real time—within 300 milliseconds—and takes action based on predefined customer preferences. This could involve blocking high-risk content immediately or slowing its distribution pending human review.
Currently, Moonbounce serves three key sectors: platforms hosting user-generated content (e.g., dating apps), AI companies creating virtual companions, and AI image generators. Its client roster includes Channel AI, Civitai, Dippy AI, and Moescape, with its system processing over 40 million daily reviews and supporting more than 100 million daily active users.
Levenson emphasized that safety can be a competitive advantage. “It’s always been an afterthought,” he said. “But our customers are finding innovative ways to make safety a part of their product story.”
The Growing Need for Guardrails
The stakes have never been higher. Recent incidents involving AI chatbots have highlighted the risks of insufficient moderation. For example, a Florida teenager tragically died in 2024 after becoming obsessed with a Character AI chatbot, prompting calls for stricter oversight. Similarly, image generators like xAI’s Grok have been misused to create nonconsensual nude imagery, sparking outrage and legal challenges.
Lenny Pruss, general partner at Amplify Partners, noted the urgency of Moonbounce’s mission. “Content moderation has always been a challenge, but with LLMs at the heart of every application, it’s even more daunting,” he said. “We envision a world where real-time guardrails enable every AI-mediated application.”
Moonbounce’s technology offers a unique advantage by operating as a third-party intermediary. “The chatbot itself has to remember thousands of tokens from previous interactions,” Levenson explained. “Our system focuses solely on enforcing rules at runtime.”
Iterative Steering: A Proactive Approach
Looking ahead, Moonbounce is developing a feature called “iterative steering” to address cases where bluntly rejecting harmful content isn’t enough. Instead of shutting down conversations about sensitive topics, the system would intercept and redirect them, modifying prompts to steer chatbots toward supportive responses.
“We want to force chatbots to be not just empathetic listeners but helpful listeners in those situations,” Levenson said.
A Full-Circle Moment?
As Moonbounce gains traction, speculation swirls about its future—including the possibility of acquisition by tech giants like Meta. While Levenson acknowledges the fit with his former employer, he remains cautious.
“My investors would kill me for saying this, but I’d hate to see someone buy us and restrict the technology,” he said. “It’s important that this benefits everyone.”
Balancing Innovation and Responsibility
Moonbounce’s rise reflects a broader imperative in the tech industry: to balance rapid innovation with accountability. As AI continues to permeate daily life, companies must prioritize safety—not as an afterthought but as a core feature. While challenges remain, Moonbounce’s “policy as code” approach offers a promising path forward, demonstrating that technology can be both powerful and responsible.
In a world increasingly shaped by AI, the question isn’t whether companies can innovate—it’s whether they can innovate safely. Moonbounce is betting that with the right guardrails, they can.
