White Circle, an AI monitoring and control startup, has raised £8 million in seed funding from investors including Romain Huet (OpenAI); Dirk Kingma (ex-OpenAI, now Anthropic); Guillaume Lample (Mistral); Thomas Wolf (Hugging Face); Olivier Pomel (Datadog); François Chollet (Keras); Mehdi Ghissassi (ex-DeepMind); Paige Bailey (DeepMind); and David Cramer (Sentry). The platform helps companies monitor, secure and improve AI models in real time by checking inputs and outputs against custom policies and identifying issues including hallucinations, prompt injection attacks and abusive users. Funding will be used to accelerate product development, expand the team across the US, UK and Europe and grow White Circle’s customer base.
Founded by Denis Shilov, White Circle was created after Shilov went viral in 2024 for bypassing the safety filters of major AI models with a single prompt. The jailbreak exposed ways to extract dangerous or restricted information from models including ChatGPT and Claude. After the post received 1.4 million views, Anthropic invited Shilov into its bug bounty programme before White Circle was built to meet demand from AI providers and enterprises looking for more control over production AI systems.
Running through a single API, the platform’s proprietary models monitor AI behaviour in real time and enforce company-defined policies. White Circle says the software can identify sensitive data leaks, block malicious AI agent instructions, detect model drift and flag users attempting to abuse systems. The platform also uses labelled feedback to improve performance over time and supports more than 150 languages. Customers include Lovable and two unnamed digital banks.
Alongside the platform, White Circle publishes research into AI model behaviour and moderation systems. In 2025, the business released CircleGuardBench, a benchmark designed to test how AI moderation models perform under real-world conditions. More recently, White Circle published KillBench, a study covering more than one million experiments across 15 AI models from providers including OpenAI, Google, Anthropic and xAI. According to the research, the models showed different biases depending on factors including nationality, religion, body type and phone brand.
AI is moving faster than our ability to guide it. We already trust it with decisions that touch millions of people – from hiring to healthcare, finance and security – and with the rise of vibe coding, anyone can ship an AI product without knowing what that model is actually doing once it's live. Until now there’s not been a platform purpose-built to monitor AI’s behaviour, catch it when it goes wrong, or shape how it acts. With White Circle, we’re finally giving companies everything they need to hold their AI accountable and optimise their models in a single place, without sacrificing security, compliance or risk.
We wanted any team, technical or not, to be able to see exactly what their AI is doing in one place – whether they're getting visibility into their models for the first time or running AI at scale and need to catch problems fast. Everything from real-time flagging to analytics and optimisation lives in one place, connected via a single API.
Denis and the White Circle team have an unusual combination of deep technical credibility and a clear commercial instinct. The KillBench research alone shows what's possible when you approach AI safety empirically rather than ideologically and the team is building the infrastructure the industry genuinely needs.
