Vstupy AI
- Git / CI/CD
- Chat / aplikace
- Prohlížeč
- CRM / ERP / Helpdesk
- Slack / Teams
- IDe / vývoj
Uživatelé a systémy posílají dotazy, dokumenty, kód a data do AI
AI Gateway
AI Gateway: Take Control of Company AI
Your people use ChatGPT, Claude, or Gemini every day. But do you know exactly what data they send into them and how much it really costs you? AI Gateway turns shadow AI into a fully auditable system. It works as an intelligent firewall that protects your know-how, controls budgets, and gives company leadership confidence that AI is used safely and according to the rules.
Zachyceno
- Rodné číslo
- API klíč
- xxxxxxx
AI Gateway
Bezpečnost, governance, audit, náklady
- Kontrola citlivých dat (CLP)
- Detekce jailbreaků a prompt injection
- Povolené modely a přístupy
- Logování a auditní stopa
- Limity a rozpočty
- Schválení rizikových akcí
Gateway zachytí škodlivé nebo nevhodné vstupy, aplikuje pravidla, zaznamená incidenty a vrací bezpečný výsledek.
AI modely
Gateway pošle dál jen bezpečný a povolený požadavek
Bezpečný výstup
- Kontrola odpovědi
- Maskování citlivých údajů
- Záznam incidentů
- Bezpečné doručení uživateli
Firma má nad AI
plnou kontrolu
plnou kontrolu
Firma ví, co bylo zablokováno, proč se to stalo a jak pravidla dál zlepšit.
Co z toho firma má
Shadow AI risks
The hidden risk growing
behind your back
Artificial intelligence is often used in companies without oversight. Employees experiment, send sensitive data to public models, and the company loses visibility into costs and security.
4
'%20fill='white'/%3e%3ccircle%20cx='47.8918'%20cy='55.4989'%20r='3.02584'%20transform='rotate(45%2047.8918%2055.4989)'%20fill='white'/%3e%3cpath%20d='M33.5101%2028.0405L29.4092%2023.9396L31.9463%2014.4712'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3cpath%20d='M60.9683%2055.4987L65.0692%2059.5996L74.5376%2057.0625'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3cpath%20d='M27.5071%2034.0432L22.4553%2028.9914L9.85547%2028.6348'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3cpath%20d='M54.9646%2061.5022L60.0165%2066.554L60.373%2079.1538'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3cpath%20d='M19.6609%2041.8881L15.4412%2037.6684L10.9245%2042.1855L1.05859%2042.1855'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3cpath%20d='M47.0005%2069.2277L51.2203%2073.4475L46.7031%2077.9642L46.7031%2087.8301'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3ccircle%20cx='31.0137'%20cy='43.6127'%20r='3.02584'%20transform='rotate(45%2031.0137%2043.6127)'%20fill='white'/%3e%3cpath%20d='M69.1686%2027.6839C69.1686%2027.6839%2074.5546%2024.5277%2078.3808%2024.2962C81.9341%2024.0813%2087.2957%2026.1981%2087.2957%2026.1981'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3cpath%20d='M61.4425%2019.9574C61.4425%2019.9574%2064.4069%2014.5549%2064.8302%2010.7452C65.3651%205.93112%2062.9284%201.83023%2062.9284%201.83023'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3c/svg%3e)
1 in 5 data leaks goes through AI
A full 20% of all corporate security incidents over the past year demonstrably involved misuse or uncontrolled use of Shadow AI.
Programs.com: Cybersecurity Education & Careers'%3e%3cpath%20d='M4%2077L15.4714%2059.4882C15.8379%2058.9288%2016.6396%2058.8798%2017.0714%2059.3904L28.3404%2072.7131C28.778%2073.2305%2029.5927%2073.1719%2029.9519%2072.5973L51.2316%2038.549C51.6053%2037.9511%2052.4639%2037.9173%2052.8834%2038.4841L62.5072%2051.4872C62.9323%2052.0617%2063.8054%2052.0176%2064.1705%2051.4033L87%2013'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-linecap='round'/%3e%3cpath%20d='M46.2543%2042.499L50.1098%2043.6996C55.474%2045.4148%2060%2047.4158%2060%2053.9335C60%2060.6798%2054.9711%2064.6819%2046.422%2064.6819H46.2543V72H44.2428V64.5676C36.3083%2063.8815%2031%2059.3077%2031%2052.447H33.6262C33.738%2057.9927%2037.9846%2061.5374%2044.2428%2062.2235V44.5L42.1753%2043.8711C38.9904%2042.8992%2032.1175%2041.2983%2032.1175%2033.9802C32.1175%2028.3202%2036.5318%2023.9751%2044.2428%2023.8607V17H46.2543V23.9179C53.5183%2024.5468%2057.9884%2029.1206%2058.4355%2036.21H55.8651C55.3064%2030.4927%2051.9538%2026.948%2046.2543%2026.3191V42.499ZM43.0694%2041.4699L44.2428%2041.8129V26.2048C38.0405%2026.262%2034.6879%2029.578%2034.6879%2033.7516C34.6879%2039.2973%2040.052%2040.4979%2043.0694%2041.4699ZM46.5896%2062.3378C53.4066%2062.3378%2057.3738%2059.1933%2057.3738%2054.0478C57.3738%2049.0166%2053.7418%2047.473%2049.4952%2046.158L46.2543%2045.1289V62.3378H46.5896Z'%20fill='white'/%3e%3cpath%20d='M20%201H70C80.4934%201%2089%209.50659%2089%2020V70C89%2080.4934%2080.4934%2089%2070%2089H20C9.50659%2089%201%2080.4934%201%2070V20C1%209.50659%209.50659%201%2020%201Z'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3c/g%3e%3cdefs%3e%3cclipPath%20id='clip0_ico_dollar'%3e%3crect%20width='90'%20height='90'%20fill='white'/%3e%3c/clipPath%3e%3c/defs%3e%3c/svg%3e)
An expensive oversight (+ USD 670,000)
According to IBM analysis, incidents in companies with high levels of uncontrolled AI cost USD 670,000 more on average. Investigating incidents without central logs is extremely complex.
IBM, August 2025'%3e%3cpath%20d='M77%2064V80.1192C77%2081.1579%2076.1698%2082%2075.1457%2082H2.8543C1.8302%2082%201%2081.1579%201%2080.1192V25.8808C1%2024.8421%201.8302%2024%202.8543%2024H15.127'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3cpath%20d='M15.8545%209H88.1455C88.6203%209%2089%209.38311%2089%209.84863V63.1514C89%2063.6169%2088.6203%2064%2088.1455%2064H15.8545C15.3797%2064%2015%2063.6169%2015%2063.1514V9.84863C15%209.38311%2015.3797%209%2015.8545%209Z'%20stroke='white'%20stroke-opacity='0.25'%20stroke-width='2'%20stroke-miterlimit='10'/%3e%3cpath%20d='M36%2041C38.7614%2041%2041%2038.7614%2041%2036C41%2033.2386%2038.7614%2031%2036%2031C33.2386%2031%2031%2033.2386%2031%2036C31%2038.7614%2033.2386%2041%2036%2041Z'%20fill='white'/%3e%3cpath%20d='M52%2041C54.7614%2041%2057%2038.7614%2057%2036C57%2033.2386%2054.7614%2031%2052%2031C49.2386%2031%2047%2033.2386%2047%2036C47%2038.7614%2049.2386%2041%2052%2041Z'%20fill='white'/%3e%3cpath%20d='M68%2041C70.7614%2041%2073%2038.7614%2073%2036C73%2033.2386%2070.7614%2031%2068%2031C65.2386%2031%2063%2033.2386%2063%2036C63%2038.7614%2065.2386%2041%2068%2041Z'%20fill='white'/%3e%3c/g%3e%3cdefs%3e%3cclipPath%20id='clip0_4058_3250'%3e%3crect%20width='90'%20height='90'%20fill='white'/%3e%3c/clipPath%3e%3c/defs%3e%3c/svg%3e)
Massive copy-paste culture
Research shows that 77% of employees routinely copy company data directly into public GenAI tools. More than half of those pastes contain sensitive internal information or proprietary code.
uSecure
Lack of control (97%)
A full 97% of AI-related incidents happened in companies that did not have properly configured access roles and rules for using language models.
Programs.com: Cybersecurity Education & CareersWe</>
Technical capabilities
How AI Gateway works in practice
AI Gateway (AI Proxy) is a central infrastructure layer between your people and external LLM models. It works as an intelligent router and firewall.
Data protection (DLP)
It automatically detects and blocks leaks of birth numbers, IDs, IBANs, source code, salaries, or business contracts. It protects against jailbreaks.
Model governance
It assigns models by role, such as Claude for developers and GPT for legal teams. It enforces that EU citizens' data cannot leave the EU and blocks risky models.
Audit & NIS2 compliance
It logs the full trail: who used which model, when, and what answer they received. It generates solid evidence for ISO 27001 and NIS2.
Cost control (FinOps)
It sets budget ceilings for teams and projects. Routine queries are automatically switched to cheaper models, while expensive ones are reserved for complex tasks.
Typical scenarios
When will AI Gateway save your skin?
Instead of complex theory, here are 4 concrete situations AI Gateway solves for you:
4
- 01A developer copies code into a public AIThey want to speed up their work, but unknowingly share internal know-how.Gateway recognizes it: It blocks the request, anonymizes it, or safely redirects it to an approved company model.
- 02Sensitive data shared through a public linkAn employee generates a public link to an AI conversation containing financial margins or NDA content.Gateway intervenes: It blocks external sharing of links with sensitive content across the board.
- 03An attempt to extract internal data (jailbreak)An attacker tries to force your company chatbot through a prompt to reveal a customer database or system instructions.Gateway stops the attack: It detects prompt injection and blocks the attempt.
- 04The token bill goes through the roofA forgotten script or careless developer generates 10 million tokens in a single day.Gateway protects you: It sends an immediate notification and prevents financial bleeding.
Target teams
Who is AI Gateway essential for?
The solution is designed for companies that need to combine AI flexibility with uncompromising security, cost control, and regulatory compliance.
~
Banks and financial institutions
Corporations and companies with over 500 employees
Public administration and healthcare
Manufacturing companies and organizations with their own IT department
Are you a CISO?
You gain peace of mind. Gateway works like a firewall that automatically filters outgoing data (personal IDs, source code, contracts) and blocks prompt injection attacks.
Are you a CIO?
You gain control. You will know exactly how much each application or department spends on AI and get complete logs for an upcoming NIS2 audit.
Are you a CTO?
You gain flexibility. Your people can safely use the best models on the market (Claude, GPT, Gemini), while you manage them centrally through one API.
Cooperation process
What cooperation looks like
We do not deploy technology blindly. The whole process is designed so you know exactly what happens at every stage and can be confident the solution matches your needs.
- 01DiscoveryTogether we identify the processes where your company faces the greatest security risks and design the first set of rules.
- 02Architecture & Proof of Concept (PoC)We design the technical connection of Gateway to your existing infrastructure and prepare an implementation plan.
- 03Implementation & IntegrationWe connect Gateway to your real systems and use cases. We test operations and deploy production security and cost rules.
- 04Adoption & ImprovementWe teach your IT and security team to work effectively with Gateway. Based on real production data, we continuously tune the system.
Trust and security
Why trust Unnits with AI governance?
We do not sell boxed products or write theoretical reports. We design and integrate infrastructure so technology works humanly, fairly, and with measurable impact on your business.
Uncompromising protection of sensitive company data.
Clear, centralized management of access and models.
Completed audits and compliance with ISO 27001 and NIS2.
Absolute control over token costs.
How secure are your AI projects?
We will gladly sit down over coffee and spend 15 minutes discussing the key security risks of your AI solutions. Specific, no fluff.
