The Facebook Insider Building Content Moderation for the AI Era
When Brett Levenson departed Apple to join Facebook in 2019, the company was grappling with intense scrutiny from the Cambridge Analytica scandal. Levenson soon realized that content moderation challenges went far beyond technology—human reviewers hurriedly interpreted lengthy policies, often inaccurately assessing flagged content long after issues arose.
From Frustration to Innovation
Recognizing the limitations of slow, manual moderation, especially with the explosion of AI-powered content, Levenson conceived the concept of "policy as code." This approach turns policy documents into dynamic software that enforces rules in real-time. This vision led to Moonbounce, which has just raised $12 million in a round co-led by Amplify Partners and StepStone Group.
How Moonbounce Works
Moonbounce integrates with companies to deliver safety and compliance wherever content is generated, whether by users or AI. Its proprietary language model reviews and enforces policies in under 300 milliseconds, taking actions like holding, blocking, or further reviewing risky content based on the company’s requirements.
Current customers include dating platforms, AI character creators, and AI image generators. Moonbounce currently supports more than 40 million daily reviews and serves over 100 million active users across the globe.
Making Safety a Product Feature
Levenson emphasizes that embedding safety directly into products can be a unique selling point, not just a backend responsibility. Major platforms, like Tinder, have seen dramatic accuracy improvements by adopting similar LLM-based solutions.
The Growing Stakes of AI Moderation
As AI-driven content grows, platforms face increasing legal, reputational, and ethical challenges, including incidents involving chatbot-induced harm or the spread of nonconsensual imagery. Consequently, many companies seek third-party experts like Moonbounce to safeguard their platforms.
What Comes Next
Alongside co-founder Ash Bhardwaj, Levenson is working on "iterative steering"—technology designed to gently redirect risky AI chatbot conversations toward positive outcomes, rather than outright refusal, enhancing support for vulnerable users.
While Levenson acknowledges that Moonbounce could complement major tech ecosystems like Meta's, he is committed to making the technology widely accessible rather than restricted to a single company.
For more detailed insights, read the full article on TechCrunch.
