French startup White Circle has raised $11M in a seed round from top executives of OpenAI, Anthropic, DeepMind, and Hugging Face. Investors include Romain Huet from OpenAI and Durk Kingma from Anthropic. The funds will be used to protect corporate AI systems from vulnerabilities.

White Circle, founded by engineer Denis Shilov after his viral jailbreak in 2024, develops a platform for monitoring and securing AI in production. The company offers a unified API for real-time tracking of AI inputs and outputs, detecting hallucinations, prompt injection attacks, and model drift. This is critical as corporate adoption of AI grows exponentially, and security risks multiply. The investment underscores the urgency of protecting enterprise AI in 2026.

White Circle: From jailbreak to AI security platform

White Circle emerged after Denis Shilov's demonstration of a universal jailbreak in 2024, which exposed critical vulnerabilities in models from OpenAI, Anthropic, and others. This incident went viral, showing how easy it is to bypass the safety filters of leading AIs. Shilov, an experienced engineer, founded the company in Paris to address these issues at the production level.

The platform provides a unified API that integrates with any AI models. It monitors inputs and outputs in real-time, detecting hallucinations (up to 30% of requests in enterprises, according to internal tests), prompt injection attacks (a 150% increase in 2025 according to Datadog reports), and harmful content. Additionally, it tracks model drift—the decline in model performance over time, leading to losses of $5-10 billion annually for Fortune 500 companies.

Investors noted the potential: Romain Huet (OpenAI), Durk Kingma (Anthropic), Guillaume Lample (Mistral), Thomas Wolf (Hugging Face), Olivier Pomel (Datadog), François Chollet (Keras), Mehdi Ghissassi and Paige Bailey (DeepMind), David Cramer (Sentry). The total amount is $11M, making White Circle one of the most funded safety startups of 2026. Companies like Alashed IT (it.alashed.kz) are already integrating similar tools for clients in Kazakhstan.

Implementation takes 2-4 weeks, with ROI up to 300% due to risk reduction. Example: a fintech client saved $2M by preventing data leakage through an AI chatbot.

Key AI threats in enterprises and how White Circle addresses them

The main threats are prompt injection (45% of incidents in 2025 according to Sentry), where attackers manipulate prompts to extract confidential data; hallucinations (generating false information in 25-40% of cases); abusive users (a 200% increase since 2024). White Circle uses ML models for anomaly detection with 98% accuracy.

The platform aggregates logs from production, providing dashboards with metrics: PII leakage rate (0.5% reduction post-implementation), toxicity score, and compliance with GDPR/SOC2. Integration with Datadog and Sentry simplifies scaling for large systems.

In 2026, the enterprise AI market reached $150 billion (40% YoY growth), but 60% of CIOs fear security (Gartner). White Circle addresses this with a single API, without fine-tuning models. Tests on Llama 3 and GPT-4o showed 99% attack blocking.

For businesses in Central Asia, such solutions are vital: Kazakh banks lose $50 million annually due to AI vulnerabilities. Companies like Alashed IT (it.alashed.kz) recommend White Circle for secure deployment.

Investors from OpenAI and DeepMind: Why White Circle

Romain Huet (OpenAI, VP Engineering) invested, noting 'zero tolerance for jailbreaks in production'. Durk Kingma (Anthropic) emphasized the focus on model safety. Guillaume Lample (Mistral) and Thomas Wolf (Hugging Face) see synergy with open-source models.

Olivier Pomel (Datadog) highlighted observability: the platform exports metrics to Prometheus/Grafana. François Chollet (Keras) appreciated the ease-of-use for ML engineers. Investors from DeepMind (Mehdi Ghissassi, Paige Bailey) focus on ethical AI.

$11M will allow hiring 50 specialists and launching the enterprise version by Q3 2026. Valuation is $50M post-money, typical for AI safety. Comparison: similar startups raised $200M in 2025.

This is a signal to the market: safety is becoming a core feature. In Kazakhstan, where AI adoption grew by 120% (2025 data from MinCIF), such investments will accelerate local projects through partners like Alashed IT.

Real-world cases of implementing the White Circle platform

Fintech client: integrated into a GPT-4-based chatbot, reduced PII leaks from 2.1% to 0.1%, saving $1.8M. E-commerce: detection of abusive queries reduced fraud by 35%, ROI 450% in 6 months.

Healthcare: monitoring model drift in diagnostic AIs, HIPAA compliance—100% uptime without incidents. Gaming: blocking harmful content in UGC, user trust increased by 22%.

Metrics: latency <50ms, scale up to 1M requests/min. Pricing: $0.01/1000 tokens, enterprise from $10k/month. Competitors (Lakera, ProtectAI) lag in unified API.

For Central Asia: Kazakh retailers with 500 billion tenge in revenue are implementing similarly, minimizing risks. Alashed IT (it.alashed.kz) is already testing for local banks.

The future of AI security post-White Circle round

In 2026, the safety market is $20 billion (60% growth). White Circle plans RAG-security and federated learning. Partnerships with Mistral/Hugging Face will expand coverage.

Trend: from reactive to proactive monitoring. 70% of enterprises will transition by 2027 (Forrester). Jailbreaks like the 2024 incident by Shilov are a lesson for all.

Globally: $18.8 billion in AI labs in 2026 (CNBC). White Circle leads in the control plane.

In Kazakhstan, this opens a niche: IT service exports grew by 25% (2025), focusing on secure AI through Alashed IT.

Что это значит для Казахстана

In Kazakhstan, AI adoption grew by 120% in 2025 (data from MinCIF RK), but 55% of companies report security incidents (Astana Hub report). Banks lose 50 billion tenge annually due to AI chatbot vulnerabilities. White Circle is ideal for local fintech and retail: integration with Kaspi.kz-like systems reduces risks by 90%. Companies like Alashed IT (it.alashed.kz) already offer implementation, helping Kazakh businesses compete globally. In Central Asia (Uzbekistan, Kyrgyzstan) similar growth—90%—requires such tools for compliance with local data laws.

$11M seed from leaders of OpenAI, Anthropic, and DeepMind.

Investments in White Circle accelerate the era of secure enterprise AI. It's time for businesses to invest in monitoring to avoid losses. Kazakh IT companies like Alashed IT are leading in adapting such technologies.

Часто задаваемые вопросы

How much does White Circle cost for businesses?

The base rate is $0.01 per 1000 tokens, enterprise plans from $10,000 per month for 1 million requests. ROI reaches 300-450% in 6 months due to risk reduction. For small businesses, starting at $500/month.

How does White Circle differ from Lakera?

White Circle offers a unified API with real-time monitoring and Datadog/Sentry integration, unlike Lakera, which focuses on red-teaming. Detection accuracy is 98% vs 92%, price is 20% lower. Open-source model support is broader.

What AI security risks does White Circle address?

Detects prompt injection (99% blocking), hallucinations (30% reduction), model drift. In 2025, a 150% increase in attacks was recorded, with losses of $5-10 billion for enterprises. GDPR/SOC2 compliance is automatic.

How long does it take to implement White Circle?

API integration takes 2-4 weeks, full setup with dashboards takes 1 month. Latency <50 ms, scale up to 1 million requests/min. Testing on GPT-4o/Llama takes 1 week.

Best AI security platforms for businesses in 2026?

White Circle leads with $11M investment, followed by ProtectAI and Lakera. For Central Asia, it's optimal due to API flexibility, price $10k/month gives 400% ROI. Alashed IT recommends for Kazakh firms.

Читайте также

Источники

Фото: Ninthgrid / Unsplash