White Circle raises $11M seed for AI control platform after viral safety exposé

Paris-based White Circle has secured $11M in seed funding from a star-studded roster of AI industry figures, including Romain Huet (OpenAI), Dirk Kingma (ex-OpenAI, now Anthropic), Guillaume Lample (Mistral), Thomas Wolf (Hugging Face), and François Chollet (Keras). The platform was founded by Denis Shilov, who made headlines in 2024 when he bypassed safety filters of ChatGPT, Claude, and other major models with a single universal jailbreak prompt—amassing 1.4M views and earning invitations to Anthropic's bug bounty program. The funding will accelerate product development and global expansion as 'vibe coding' enables anyone to ship AI products, increasing the risk of models breaking safety rails.

The White Circle platform provides a single API for real-time monitoring of AI inputs and outputs, catching harmful content, hallucinations, prompt injection attacks, model drift, and abusive users. Custom policies allow automated enforcement actions like rate limiting or banning bad actors. Specific use cases include detecting a fintech model leaking sensitive customer data, blocking an AI agent from executing malicious instructions (e.g., deleting files), and flagging rising user frustration when a model fails. The system supports 150 languages, is SOC 2 Type I/II certified, and HIPAA compliant. In May 2025, White Circle also published CircleGuardBench, a benchmark for testing AI moderation models under real-world conditions. Head of Design Elena Iumagulova emphasized that the platform makes AI behavior visible to any team, technical or not.