White Circle raises $11M seed to monitor and harden AI models

Pumpkin Intelligence Inc., operating as White Circle, has closed an $11 million seed round after founder Denis Shilov’s 2024 viral demonstration showed a single prompt could jailbreak many proprietary AI models. The funding, backed by senior AI figures, aims to accelerate tools that monitor model behavior and shore up safety gaps exposed by the demonstration. Investors named in the round include Romain Huet, Dirk Kingma, Mehdi Ghissassi and Olivier Pomel. White Circle’s raise follows broad attention from the industry: Shilov’s research highlighted systemic weaknesses that prompted responses from major model developers and increased demand for third — party guardrails.

Shilov’s tests showed attackers could bypass protections to unlock instructions for making drugs and weapons, access dangerous or illegal information and extract sensitive material such as ChatGPT’s system prompt. Those findings underscored the risk that deployed models can be repurposed to surface harmful content or leak confidential inputs unless actively monitored.

White Circle says its scanning service and API produce data teams can use to assess model performance, choose appropriate models and iteratively improve them. The company describes use cases such as detecting a fintech model leaking sensitive data and blocking exfiltration, while defensive models learn normal behavior patterns and improve via labeled feedback.

The funding is intended to scale White Circle’s detection and remediation tooling and support wider adoption among developers and enterprises that need to validate model safety before deployment. “AI is moving faster than our ability to guide it,” Shilov said, summing up the urgency behind the product. For organizations deploying models, the company positions its service as a way to audit and harden behavior that might otherwise go unnoticed.