All analysis was generated autonomously, without human review. Scores are analytical opinions drawn from the cited public sources, without hands-on testing. They are not audits, certifications, investment reports, purchasing advice, or evaluations of quality.
White Circle sells a control layer a company runs between its own application and the AI models it uses, so the company decides what those models may do. Individuals who built at OpenAI, Anthropic, and DeepMind backed the 11 million dollar seed, funding a layer that constrains the class of models their companies build. The founder earned that audience by jailbreaking major AI models with a single prompt, and early customers include the app builder Lovable and two large digital banks. White Circle is betting that safety cannot be settled inside the models, so its natural buyer is the regulated enterprise that wants to own its AI controls rather than rely on a model provider.
| Description | White Circle is an AI control platform that red-teams AI systems for failures, applies low-latency guardrails against prompt injection and unsafe output, and monitors AI inputs and outputs for harmful content, abuse, and model drift. | [f1] |
|---|---|---|
| Funding | $11M total | [f2] |
| Latest funding | Seed | [f2] |
| Deployment | SaaS | [f3] |
| Compliance | HIPAA, SOC 2 Type 2 | [f3] |
| Product | What it does |
|---|---|
| White Circle | A unified API that red-teams AI for failures, applies low-latency guardrails against prompt injection and unsafe output, and monitors AI inputs and outputs for harmful content, abuse, and drift. |
AI Defense Matrix
| Govern | Identify | Protect | Detect | Respond | Recover | |
|---|---|---|---|---|---|---|
| AI-Workload Platforms Inference servers, training platforms, vector DB platforms, and the model-loading supply chain. | ||||||
| AI Orchestration Tools Agentic orchestration tools, plus their plugins, skills, hooks, system prompts, scaffolding, harnesses, configuration settings, and MCP clients on user devices. | ||||||
| AI-Generated Code Code produced by AI tools, AI-assisted reviews, AI-generated infrastructure-as-code and tests, and vibe-coded apps that bypass CI/CD. | ||||||
| AI Gateways & Routers MCP proxies and gateways, LLM routers, outbound AI-service traffic, shadow AI egress, and model-registry traffic. | ||||||
| AI Model Model weights, fine-tuning checkpoints, model cards, registries, AIBOM, and the third-party LLMs your enterprise consumes. | ||||||
| Training Data Datasets used for training, fine-tuning, and continued learning. | ||||||
| Runtime AI Data User prompts, inference inputs, RAG content, vector DB content, persistent agent memory, and interaction history. | ||||||
| AI Agent Identities AI agents as non-human principals, plus credentials, keys, permission scopes, service accounts, and delegation chains across agents and tools. |
White Circle provides security for AI and is mapped to the AI Defense Matrix. [f4]
How well the company can compete in its security market, scored across eight dimensions against public evidence.
| Dimension | Score | Rationale |
|---|---|---|
| Problem Clarity How precisely the company defines its problem, with evidence the problem exists at the scale claimed. | 3/5 | The control-layer problem is defined clearly and the failure modes (hallucination, prompt injection) appear in press, but the quantifying evidence (the million-experiment bias study) is the company's own research rather than independent corroboration. [s2, s3] |
| Capability Depth How specific the technical capabilities are, with evidence beyond marketing claims such as docs and third-party validation. | 3/5 | Press details the unified API and the homepage lists test, protect, observe, and improve, yet CircleGuardBench and KillBench are the founder's own published benchmarks (a team signal), leaving the capability without an external product validation point. [s3, s1] |
| Market Timing Whether the market is ready for this product, with evidence that buyers are actively seeking solutions. | 3/5 | The 2024 jailbreak wave and the move of models into production are real enablers, but buyer-side demand reduces to named adoption (Lovable plus two banks) with the rest resting on the vendor's thesis and competitors expanding, short of multiple independent demand signals. [s3, s2] |
| Team Credibility Demonstrated domain expertise with public signals such as prior exits, publications, and industry recognition. | 3/5 | Founder Denis Shilov's 2024 viral jailbreak is one public event rather than a sustained record, and two published benchmarks plus operator backers (OpenAI, Anthropic, DeepMind) show recognition without a prior in-domain exit. [s2, s3] |
| GTM Proof Evidence of actual traction (customers, revenue signals, partnerships) beyond stated intentions. | 3/5 | Press reports a billion API requests and names one customer (Lovable) with the digital banks unnamed, so the line carries a single named reference, and the notable operator backers add only a small indirect-signal bump. [s3] |
| Funding Efficiency Whether funding matches go-to-market ambition, with signs of capital-efficient growth. | 3/5 | The 11 million seed is proportional to the early stage with visible shipping (a billion requests, a named customer, SOC 2 and HIPAA, two benchmarks), but no revenue or margin is disclosed, so output per dollar stays unconfirmed. [s2, s3] |
| Category Clarity Whether the company creates or fits a recognizable category that buyers can quickly place in their stack. | 3/5 | Press places White Circle in AI observability and governance alongside Arthur AI and WhyLabs, but that category is still forming and contested across guardrails, observability, and governance, so placement is credible rather than settled. [s3] |
| Incumbent Defensibility How vulnerable the core value proposition is to absorption as a feature by a platform vendor. | 3/5 | The control-and-guardrail capability is the kind cloud platforms and model providers can bundle, so the line sits level with the pure-play cluster, with its own models and telemetry flywheel the difference that keeps it from a lower score. [s3, s1] |
Enterprises are embedding AI models into products faster than they can control what those models do, and White Circle defines the problem precisely. Its founder demonstrated the gap by jailbreaking most major models with a single prompt, and the company frames the need as a control layer each business owns to decide what its AI is allowed to do.
The pains are concrete and current. Independent press describes the platform detecting hallucinations, prompt injection attacks, harmful content, model drift, and abusive users in real time, the failure modes that turn a deployed model into a liability for the team that shipped it.
The buyer is the company putting an AI feature into production, including regulated ones. White Circle's own research, a study of bias in AI decision-making across more than a million experiments, evidences that the problem persists across every major model, which is the demand the product is built to serve. [s2, s3, s1]
White Circle delivers four capabilities through one interface: red-teaming to surface failures, low-latency guardrails to block unsafe input and output, observability to analyze behavior, and optimization through dynamic prompts and model routing. The homepage presents these as test, protect, observe, and improve.
The protection layer is the security core. Press describes a unified API that monitors AI inputs and outputs in real time to detect hallucinations, prompt injection, harmful content, and model drift, with custom enforcement policies, automated moderation, and support for more than 150 languages.
The product is backed by the company's own specialized models and research. White Circle published CircleGuardBench, a moderation benchmark, and KillBench, a large-scale bias study, technical output that validates the detection work beyond a marketing claim. [s1, s3, s2, s4, s5]
White Circle positions itself in the emerging AI observability and governance market as a vendor-neutral control layer. Press places it alongside Arthur AI and WhyLabs, with larger monitoring vendors such as Datadog also expanding into AI monitoring, so buyers can locate the category without vendor coaching.
Its structural stance is deliberately between the enterprise and the model maker. Because it controls models from OpenAI, Anthropic, Mistral, and others from the outside, it competes less with the labs than with other independent control and guardrail vendors and with the native safeguards the platforms bundle.
Against the guardrail specialists such as Lakera and Prompt Security, White Circle differentiates on breadth, pairing runtime protection with red-teaming, observability, and optimization rather than a single screening function. [s3, s1, s6]
White Circle shows unusual traction for its stage. Press reports its infrastructure has processed more than one billion API requests and that customers include the application builder Lovable and two large digital banks, named and unnamed references that evidence real adoption.
The go-to-market motion is developer-led through a single API, widened by the founder's public credibility. The viral jailbreak that started the company gave it standing with the AI labs, several of whose leaders then backed the seed round, a distribution advantage into the AI ecosystem.
Compliance readiness supports the enterprise motion. The company carries SOC 2 Type I and II certification and is HIPAA compliant, the posture the regulated banking and healthcare buyers it targets require before deploying. [s3, s2]
White Circle's credibility starts with its founder. Denis Shilov went viral in 2024 by jailbreaking most major models with a single prompt, work that drew the attention of the labs whose models he broke and established adversarial-AI standing the company was built on.
The backer roster is a strong external signal. Press reports the seed round drew operators from OpenAI, Anthropic, and DeepMind, including OpenAI's head of developer experience and Datadog's chief executive, who chose to fund a company that controls their own and rivals' models.
The research output reinforces the technical claim. CircleGuardBench and KillBench are published studies that demonstrate applied capability in AI safety measurement, more than a founding pedigree asserted on a page. [s2, s3, s6]
White Circle is in production and used at scale, with press reporting more than a billion API requests processed and named enterprise customers, so the platform is past the prototype stage despite the company's seed age.
Trust posture is built for regulated buyers. The company holds SOC 2 Type I and II certification, is HIPAA compliant, and supports more than 150 languages, the controls a bank or healthcare buyer reviews before adopting a safety layer.
The open trust question is independent verification of efficacy. White Circle's own benchmarks demonstrate rigor, but the public record offers no third-party evaluation of its detection quality against alternatives, so a buyer weighs the company's research against an outside test it does not yet have. [s3, s1]
| Company | Relationship | Note | Compare |
|---|---|---|---|
| Arthur AI | competes with | An AI observability and governance vendor named alongside White Circle in the same market. | |
| Lakera | competes with | A focused guardrail layer for prompts, outputs, and agent tool calls. | |
| Prompt Security | competes with | A runtime safeguard across LLM providers and self-hosted models. | |
| Datadog | competes with | A monitoring vendor expanding into AI observability, named as an adjacent competitor. | N/AWe scored these companies at different scopes, so the totals measure different things. |
| WhyLabs | competes with | An AI observability vendor placed in the same governance market. |
Add analyzed competitors to compare them side by side with White Circle.
A closer look at the company's product strategy, measuring how defensible it is against market forces and examining the eight areas behind it.
pivot urgently
White Circle builds durability from accumulation rather than position. Press reports a billion-plus processed requests and, separately, detection models improving through labelled feedback, activity showing where a telemetry asset could form without naming one, and it carries SOC 2 Type II and HIPAA certification suited to its bank customers and healthcare prospects. What it lacks is hard lock-in. The product is an API a customer integrates and can replace, switching cost is the tuning and history built up rather than a migration barrier, and the guardrail capability is the kind cloud platforms could bundle. Its durable assets are the learning loop and the compliance posture, both of which a well-funded rival could match over time, so its edge is being early, not a lasting lock.
| Dimension | Score | Rationale |
|---|---|---|
| Value Delivery Does the product sell software as the product, or judgment, trust, or accountability with software as the delivery mechanism. | 1/3 | Customers integrate and operate a control API, a software product they configure and run, rather than a service in which White Circle accepts judgment or accountability for the safety outcome. |
| Switching Cost How expensive leaving is for a customer: data portability, integrations, learned workflows, network effects, regulatory data residency. | 2/3 | The API runs inline with custom policies and a model that learns from labelled feedback over time, so leaving means re-integrating and re-tuning elsewhere, friction the record supports without documenting whether accumulated tuning is portable, short of a hard migration barrier. |
| Compliance Moat Whether certifications, liability acceptance, or audit trails block an easy replacement. | 1/3 | White Circle carries SOC 2 Type I and II certification and HIPAA compliance, credentials its named banking customers expect and its HIPAA posture extends toward healthcare. These are commercial, table-stakes attestations, with no evidenced federal authorization behind this specific product, so a determined rival could clear the same bars. |
| Problem Complexity Whether the product requires ML, optimization, real-time systems, or years of specialized expertise. | 3/3 | Real-time detection of hallucination, injection, and drift across 150 languages on the company's own models, plus published bias and moderation research, sits in machine-learning territory that takes specialized expertise to build. |
| Buyer Profile Whether buyers are SMB operators, mid-market IT teams, or regulated enterprises and governments with procurement gates. | 2/3 | The primary motion is a developer-integrated API. The one named customer is the app builder Lovable, while the cited press adds two unnamed digital banks, so the record gestures at a regulated-enterprise buyer without identifying one or establishing its class. That evidences mid-market governance rather than a demonstrated regulated-enterprise buyer base. |
| Layer Whether the product is an end-user application, a platform with application features, or infrastructure other applications depend on. | 2/3 | White Circle is an API and platform an application integrates and can swap, a control layer beside the app rather than infrastructure other software cannot replace. |
| Proprietary Data, Content, or IP Whether the product accumulates datasets, content licenses, or IP that a rival cannot recreate from scratch. | 1/3 | Press reports a billion-plus processed requests and, separately, models improving through labelled feedback, but no named non-public dataset or verified cross-customer corpus appears in the record, so no durable data asset is evidenced yet. |
White Circle targets the company putting an AI model into production and needing to control what it does, a buyer that spans the developer shipping an app and the regulated enterprise deploying a customer-facing assistant. The single API is the entry point for both.
The evidenced segments skew to demanding buyers. Press names the app builder Lovable and two large digital banks as customers, and the company's SOC 2 and HIPAA posture targets banking and healthcare, segments where AI output carries direct liability.
The segmentation follows the architecture. Because the control layer sits outside the model, its addressable buyer is not tied to one provider, though the cited pages stop short of documenting provider-agnostic support, which widens the market but also places it against both neutral rivals and the model platforms' own controls.
White Circle pairs four functions in one platform: red-teaming to surface failures, low-latency guardrails to block unsafe input and output, observability to analyze behavior, and optimization through dynamic prompts and model routing. The protection layer is the security core.
The detection runs on the company's own models. Press describes a unified API monitoring inputs and outputs in real time to catch hallucinations, prompt injection, harmful content, and drift, with custom enforcement policies and support for more than 150 languages, and the platform learns from labelled feedback over time.
Published research backs the capability. CircleGuardBench measures moderation performance under real-world conditions and KillBench studies bias across more than a million experiments, technical output that evidences depth beyond a feature list.
Go-to-market centers on a single API developers integrate, amplified by the founder's public credibility, with both pricing tiers routed through access requests. The 2024 viral jailbreak gave White Circle standing with the AI labs, several of whose leaders then joined the seed round, an unusual distribution advantage into the AI ecosystem.
Traction is real for the stage. Press reports more than a billion API requests processed and customers including Lovable and two large digital banks, named and unnamed references rather than design partners alone.
Compliance underwrites the enterprise motion. SOC 2 Type I and II certification and HIPAA compliance equip White Circle for the regulated banking and healthcare buyers that expect those attestations in procurement.
White Circle publishes no price. Its pricing page lists two access-request plans, a Startup tier with a Request Access call to action and an Enterprise tier routed to Contact Us, so a buyer cannot see a number without engaging sales.
The billing mechanism is not stated. The product is delivered as an API and the company reports more than a billion requests processed, which makes a usage-based model plausible, but the pricing page asserts no metered or per-request billing and the Enterprise tier advertises unlimited usage, so the unit of charge is unverified rather than a stated consumption fee.
The two-tier, contact-sales structure fits an early enterprise motion. White Circle sells through direct access requests rather than transparent self-serve pricing, which suits the regulated banking and healthcare deals its compliance posture targets, though the published Startup tier offers a Request Access path rather than a public price or self-serve purchase.
White Circle is delivered as an API the customer integrates into its application request flow, with on-premises deployment listed at the Enterprise tier, so where the detection infrastructure runs varies by tier while the customer configures policy. The platform operates inline in real time on inputs and outputs.
Operations support enterprise control. Press describes custom enforcement policies, automated moderation actions such as rate limiting or banning users, and analysis of model performance across use cases, with support for more than 150 languages.
Adopting the API concentrates reliance on White Circle as the control plane, the Enterprise on-premises option moves operation into the customer's environment, and the cited pages do not detail who owns scaling and model upkeep in either form.
Trust rests on certification, scale, and research. White Circle holds SOC 2 Type I and II certification and is HIPAA compliant, the posture regulated buyers require, and press reports more than a billion requests processed, evidence the platform operates at production scale.
The founder's credibility and the backer roster reinforce it. The 2024 jailbreak that exposed model weaknesses, followed by a seed round from operators at OpenAI, Anthropic, and DeepMind, signals that serious figures vouch for the approach.
The open trust gap is independent efficacy testing. White Circle's own CircleGuardBench and KillBench demonstrate rigor, but the record offers no third-party benchmark of its detection against rivals, so a cautious buyer weighs the company's research against an outside test it does not yet have.
White Circle is built to be the control plane an enterprise places over the models it uses, a position outside the model rather than inside one vendor's stack. It sits between the application and the model as a unified control API, and the reviewed record does not document named provider integrations.
Outward, the platform routes and optimizes across models through dynamic prompts and model routing, so it positions itself as the layer that chooses and constrains models rather than competing to build them, and individuals from the model labs are among its backers.
Inward, adopting White Circle concentrates AI safety operations on its API and models. The dependence is the same concentration question every control-layer vendor raises, traded for the convenience of one control point in front of the models a company uses.
White Circle is led by founder and chief executive Denis Shilov, whose 2024 viral jailbreak of most major models established adversarial-AI credibility and drew the attention of the labs whose models he broke. That origin is the team's defining signal.
The backer roster extends the credibility. Press reports the seed drew operators from OpenAI, Anthropic, and DeepMind, including OpenAI's head of developer experience and Datadog's chief executive, a set that vouches for both technical depth and commercial instinct.
The research arm demonstrates applied capability. CircleGuardBench and KillBench are published studies in AI safety measurement, evidence the team produces technical work rather than a pedigree asserted on a page, though the company remains early and engineer-heavy by its own account.
| Id | Source | Tier | Accessed |
|---|---|---|---|
| f1 | White Circle homepage | official | 2026-06-15 |
| f2 | SiliconANGLE: White Circle raises 11M to secure and monitor AI model behavior | press | 2026-06-15 |
| f3 | AI Defense Matrix Catalog entry | other | 2026-06-14 |
| f4 | AI Defense Matrix Catalog mapping | other | 2026-06-15 |
| Id | Source | Tier | Accessed |
|---|---|---|---|
| s1 | White Circle homepage “Test Automatically red team your AI to find every failure Protect Safeguard your app with custom low-latency guardrails Observe Analyze user behavior with clustering and custom metrics Improve Optimize your AI with dynamic prompts and model routing” | official | 2026-06-15 |
| s2 | SiliconANGLE: White Circle raises 11M to secure and monitor AI model behavior “Artificial intelligence guardrail and monitoring startup Pumpkin Intelligence Inc., which operates as White Circle, announced today it raised $11 million in seed funding from a who's who of AI leadership after its founder showed it was possible to break most models with a single prompt.” | press | 2026-06-15 |
| s3 | TechFundingNews: French startup White Circle secures 11M from OpenAI, Anthropic, DeepMind leaders “its infrastructure has already processed more than one billion API requests and is currently used by customers, including Lovable and two of the world's largest digital banks. White Circle is also SOC 2 Type I and II certified and HIPAA compliant.” | press | 2026-06-15 |
| s4 | Fortune: White Circle raises 11 million to stop AI models from going rogue “White Circle's main product is a real-time enforcement layer for AI applications. If a user tries to generate malware, scams, or other prohibited content, the system can flag or block the request.” | press | 2026-06-30 |
| s5 | SecurityWeek: White Circle raises 11 million for AI control platform “The AI control platform monitors all AI inputs and outputs to identify risks such as hallucinations, harmful content, prompt injection attacks, abuse, and model drift.” | press | 2026-06-30 |
| s6 | Tech.eu: White Circle lands 11M to help companies secure AI systems “The AI governance platform for monitoring and securing AI systems in real time raised funding from leaders at OpenAI, Anthropic, DeepMind, Hugging Face and Mistral after its founder gained attention for jailbreaking major AI models.” | press | 2026-06-30 |
| Id | Source | Tier | Accessed |
|---|---|---|---|
| s1 | White Circle homepage “Test Automatically red team your AI to find every failure Protect Safeguard your app with custom low-latency guardrails Observe Analyze user behavior with clustering and custom metrics Improve Optimize your AI with dynamic prompts and model routing” | official | 2026-06-18 |
| s2 | SiliconANGLE: White Circle raises 11M to secure and monitor AI model behavior “Artificial intelligence guardrail and monitoring startup Pumpkin Intelligence Inc., which operates as White Circle, announced today it raised $11 million in seed funding from a who's who of AI leadership after its founder showed it was possible to break most models with a single prompt.” | press | 2026-06-15 |
| s3 | TechFundingNews: French startup White Circle secures 11M from OpenAI, Anthropic, DeepMind leaders “used by customers, including Lovable and two of the world's largest digital banks. White Circle is also SOC 2 Type I and II certified and HIPAA compliant. White Circle has raised $11M from operators and investors across OpenAI, Anthropic, DeepMind, Hugging Face, Mistral, Datadog, and Sentry.” | press | 2026-06-18 |
| s4 | White Circle pricing page “Pricing. We've got a plan for everyone, whether you're just getting started or already playing at the top. Startup: Request Access. Enterprise: Contact Us. Unlimited usage. On-premises deployment.” | official | 2026-06-18 |
| s5 | Fortune: White Circle raises 11 million to stop AI models from going rogue “White Circle's main product is a real-time enforcement layer for AI applications. If a user tries to generate malware, scams, or other prohibited content, the system can flag or block the request.” | press | 2026-06-30 |
| s6 | SecurityWeek: White Circle raises 11 million for AI control platform “The AI control platform monitors all AI inputs and outputs to identify risks such as hallucinations, harmful content, prompt injection attacks, abuse, and model drift.” | press | 2026-06-30 |
| s7 | Tech.eu: White Circle lands 11M to help companies secure AI systems “The AI governance platform for monitoring and securing AI systems in real time raised funding from leaders at OpenAI, Anthropic, DeepMind, Hugging Face and Mistral after its founder gained attention for jailbreaking major AI models.” | press | 2026-06-30 |
This site is an experimental research aid created by Zeltser Security Corp. All its data gathering and analysis was performed autonomously without human review, and it can contain errors of fact, interpretation, and judgment that a human reviewer might catch.
The analyses are statements of opinion, not statements of fact. Machine analysis produced the scores, summaries, and matrix placements by weighing the public sources each page cites, and reasonable people can weigh the same sources differently. Where a page states a fact, it cites the public source and the date it was checked, and the statement is only as accurate as that source. Unless a profile expressly says otherwise, the analysis involves no hands-on testing and no independent validation of any company's products or services.
Nothing here is professional, security, legal, financial, investment, or purchasing advice, and nothing here is a recommendation to invest in, do business with, or avoid any company. Inclusion of a company is not an endorsement, and absence of a company is not a judgment about it. Reading this site creates no advisory or client relationship. Verify any detail you plan to act on against the vendor's current materials.
The content is provided "as is" and "as available," with all warranties disclaimed, express or implied, including merchantability, fitness for a particular purpose, accuracy, and non-infringement. No entry is warranted to be complete, current, or correct. Companies change, vendors update their claims, sources can be wrong, and automated analysis can misread them.
To the fullest extent permitted by law, the operator, Zeltser Security Corp, is not liable for any damages that arise from using this site or relying on its content, including direct, indirect, incidental, special, and consequential damages and lost profits, even if advised that such damages were possible. If you are dissatisfied with the site or disagree with these terms, your remedy is to stop using it.
Entries link to vendor pages, press coverage, and other external sites that Zeltser Security Corp does not control and is not responsible for. A link is not an affiliation with the destination or an endorsement of it. Product and company names and trademarks are the property of their owners, used here nominatively to identify the companies described. Short quotations from cited sources appear for identification and commentary.
Use, quotation, automated retrieval, and redistribution of the content are governed by the Terms of Use at cybercompanyprofiles.com/terms, which permit personal and internal business use with attribution and prohibit republication and resale.