After bypassing ChatGPT’s safeguards, WHITE CIRCLE founder Denis Shilov raises €9.35 million

White Circle, a startup specialising in the supervision and security of artificial intelligence models, has announced an $11 million funding round, approximately €9.35 million. The round brings together several major figures from the global AI ecosystem, including Romain Huet, Dirk Kingma, Guillaume Lample, Thomas Wolf, François Chollet, Olivier Pomel, and Paige Bailey.

Founded by Denis Shilov, White Circle develops a platform enabling companies to monitor the behaviour of their AI models and autonomous agents in real time. The startup claims it already processes more than one billion API requests and counts several major international groups among its clients, including two leading global digital banks as well as Lovable.

The funding round comes as companies accelerate the deployment of AI agents in critical environments, including customer relations, finance, cybersecurity, human resources, and software automation. The market is gradually shifting from model development toward operational supervision.

Before launching White Circle, Denis Shilov became known within the AI ecosystem after publishing in 2024 a “universal jailbreak” capable of bypassing the safety mechanisms of the leading generative AI models on the market. According to the company, the prompt made it possible to obtain responses normally blocked by systems developed by OpenAI or Anthropic. The safeguards designed to prevent the generation of dangerous, illegal, or sensitive content could reportedly be neutralised with a single instruction.

The publication quickly went viral, surpassing 1.4 million views. It also attracted the attention of several American AI labs. Denis Shilov was later invited to participate in Anthropic’s bug bounty programme before launching White Circle.

White Circle is building an observability and control layer designed to sit between AI models and enterprise applications. Its platform analyses model inputs and outputs in real time in order to detect hallucinations, identify prompt injections, monitor behavioural drifts, block malicious actions, and prevent sensitive data leaks.

Companies can define their own control policies to determine what is authorised or prohibited. White Circle also provides automated mechanisms for rate limiting, blocking, or banning. The system is designed to operate across multiple AI model providers through a single API. The company claims support for more than 150 languages.

Among the use cases described by White Circle, the platform can prevent an AI agent from executing destructive commands, detect abnormal behaviour within financial workflows, or identify manipulation attempts intended to bypass a model’s internal rules.

The startup’s development illustrates the rapid evolution of the generative AI market. After the race for models and GPU infrastructure, a new segment is emerging around the supervision of production AI systems. The widespread adoption of low-code tools and “vibe coding” platforms is dramatically accelerating the deployment of AI applications within enterprises. Non-specialised teams can now connect models to databases, CRMs, ERPs, or financial tools within hours.

This democratisation, however, creates a new attack surface. Companies must now manage agents capable of interacting with critical systems without always having clear visibility into their actual behaviour. The risks include the exposure of sensitive data, operational drifts, malicious manipulation, and the execution of unintended actions.

White Circle is positioning itself precisely on this operational supervision layer. The parallel with the cloud market is evident. Following the explosion of cloud infrastructure during the 2010s, companies such as Datadog or Sentry established themselves as indispensable observability layers for monitoring distributed architectures. White Circle is now attempting to occupy a comparable role for AI systems.

The startup is also seeking to strengthen its technical credibility through the publication of research on risks associated with generative models. In 2025, it published “CircleGuardBench”, a benchmark designed to evaluate the robustness of AI moderation models under real-world conditions. More recently, White Circle introduced “KillBench”, a study based on more than one million experiments conducted across fifteen models from companies including OpenAI, Google, Anthropic, and xAI.

According to the startup, this research highlighted behavioural biases linked to nationality, religion, physical appearance, and certain cultural markers. The study also claims that some structured formats used in enterprise AI integrations significantly reduce the refusal mechanisms built into models.

White Circle’s funding round also comes amid a gradual tightening of regulatory requirements surrounding artificial intelligence. Companies must now demonstrate their ability to trace model decisions, control agent actions, document behavioural drifts, and limit legal risks associated with automation.

With the emergence of AI agents capable of interacting directly with operational systems, behavioural supervision of models could become a structural component of enterprise AI architectures. White Circle is betting precisely on that evolution.