White Circle

White Circle is an AI control and safety platform whose software sits between a company's users and its AI models, checking inputs and outputs in real time against customer-defined policies. The product spans four stages of the AI lifecycle: testing AI systems for failure modes, jailbreaks and hallucinations before release; enforcing guardrails across inputs, outputs and tool calls; analyzing how users and models behave over time; and optimizing systems through prompt engineering and model routing driven by live signals.

Capabilities described by the company include blocking unsafe inputs, jailbreak and prompt-injection prevention, PII detection, output validation and instruction-following verification, data-leak and hallucination detection, tool-abuse and token-drain prevention, compliance checks, risk scoring, anomaly detection and custom metrics. Policies can be written in natural language across safety, security, compliance and reliability, varied by user location or jurisdiction, and enforced by blocking, rewriting or escalating risky behavior — for example alerting a security channel, requesting human confirmation, or terminating a session. Every decision is auditable, showing what happened and which policies were triggered. The platform is offered as an API with an accompanying command-line tool (installable via pip as whitecircle_cli), and the company maintains a public trust center and status page.

Founding story

In late 2024, Denis Shilov devised a universal jailbreak prompt that instructed leading AI models to behave like an API endpoint rather than a chatbot with safety rules, causing them to answer requests they were supposed to refuse. He posted it on X, where it went viral overnight and led to an invitation from Anthropic to privately test its models. That experience convinced him that jailbreaks were only one facet of a broader problem: companies integrating AI into workflows had few ways to control model behavior once users interacted with it, which became the premise for White Circle.

Business model

B2B SaaS. White Circle sells an API-based control layer to businesses deploying AI in production, with enterprise-oriented features such as unlimited custom policies, regional policy variation, audit logging and a trust center; the site offers self-serve sign-up alongside a sales-led demo request.

Sold as a software platform/API to business customers; a self-serve "Get Started" path and a "Contact Sales" enterprise path are both offered.

Traction

The company states its platform has processed more than one billion API requests and cites Lovable, a vibe-coding startup, along with several fintech and legal companies, as users. A customer testimonial from CISO Igor Andriushchenko states that White Circle protects millions of AI interactions per day. Its Hugging Face organization lists 11 team members, five public datasets and two spaces.

Latest developments

In May 2026 White Circle disclosed an $11 million seed round to be used for team expansion, product development and customer growth in the U.S., U.K. and Europe, and published KillBench, a study of more than one million experiments across 15 models from OpenAI, Google, Anthropic and xAI examining behavior in forced decisions about human lives. Additional research on distributed training methods, Blackwell B300 kernels and offline GRPO is announced as forthcoming.