AI application security testing: prompt injection, data leakage, agent and API abuse. Plus a pentest of the rest of the application, regression tests in CI and performance. We agree on the scope and the report date before we start, and the retest after fixes is included.
A language model executes instructions found in text. Any text that reaches it (a user message, a document in RAG, a web page, an e-mail) can contain instructions you never wrote.
Testing starts with the AI layer and covers the whole application around it: the API, permissions, tests in CI and performance.
Chatbots, assistants, agents, RAG, integrations with OpenAI/Anthropic/Google/open-source models. Prompt injection, data leakage, excessive tool permissions, API abuse and filter bypasses, mapped to the OWASP Top 10 for LLM Applications. A report with example prompts and recommendations.
The model sits inside a web application with an API, sessions and permissions. Pentest per OWASP Top 10 / ASVS, report, free retest.
A suite of prompt tests run after every change to the prompt, model or code, so a patched vulnerability does not come back. Plus e2e and accessibility tests in Playwright.
k6: how many users the application can take, how model response latency grows, whether token limits work, what breaks first.
Before a release, after a big change, or when the automation says “green” and users say otherwise.
A review of your test automation: coverage, stability, runtime, maintenance cost. A concrete list of what to fix.
You have built a chatbot, copilot or agent and are releasing it to users. The test treats it the way an attacker would, and you get a report you can show an investor or an enterprise customer.
Typical engagement: AI + API Pentest before launch
You add RAG, chatbots and LLM automations to client projects. An external QA/security engineer joins per project or on a retainer, and you hand the client an application with an independent report and a badge.
Typical engagement: AI Security Quick Check on every delivered project + evals in CI
You have a vendor bot or a no-code integration with access to orders and customer data. The test shows what the bot can reveal and do, whether it discloses that it is AI (an AI Act requirement) and whether customer data leaks (GDPR). In plain language, no jargon.
Typical engagement: AI Security Quick Check + a fix list for the vendor
This is not a law firm’s or an auditor’s service. The result is technical evidence (tests, reports, retests) that these regulations require or that an auditor wants to see.
From 2 August 2026 chatbots and assistants must clearly tell users that they are talking to AI (Art. 50). High-risk systems will also need risk management, logging and testing, but those duties were postponed to 2027–2028. The test checks the Art. 50 obligation technically and, for high-risk systems, provides evidence of robustness and cybersecurity (Art. 15); the result is documented.
Art. 32: appropriate technical and organizational measures and regular testing of their effectiveness (Art. 32(1)(d)). Personal data in prompts, in the RAG database, in model logs and at an external LLM provider: the test checks that it does not leak through responses, the API or exports.
The directive, transposed in Poland in the National Cybersecurity System Act (KSC), requires risk management and regular security testing. An application pentest (including the AI layer) with a report and retest is a concrete item in your documentation.
For fintechs and their suppliers: application security testing, including AI components, and performance under load as part of operational resilience.
If your service falls under the European Accessibility Act (e.g. an online store or banking), the chat interface must be accessible too: keyboard, screen reader, contrast. The testing can cover conformance with WCAG 2.2 AA, and this website meets it itself.
You send an e-mail, we schedule a call. You describe what we are testing and why: which model, which data, which tools the agent has. We agree on scope, environment and dates. If the call is to cover details, we sign the NDA before it.
Within 2 working days you get a quote: scope, rules (NDA, testing window, token budget, emergency contact), date and price. Testing starts after your written consent.
Testing runs on the environment you designate. Critical findings (e.g. a customer data leak) reach you immediately, not at the end.
A report with priorities, example prompts/requests, evidence and recommendations (PL or EN) + a walkthrough meeting. If the fixes are ready within 30 days of the report, you get a free retest and an updated report.
Every project is quoted individually after a free call. You get a quote with the price within 2 working days. The scope and timeline in the quote are binding, so you know what we are agreeing on.
| Package | What it includes | Time |
|---|---|---|
| AI Security Quick Checkrecommended | review of one AI application: prompt injection (direct and indirect), prompt and data leakage, tool permissions, API limits, AI Act transparency obligations; report with priorities, walkthrough, free retest | report in 5 working days |
| AI Security Review | full testing of the AI layer (mapped to the OWASP Top 10 for LLM Applications) + business logic + RAG + agents; report, walkthrough, free retest | approx. 10 working days |
| AI + API Pentest | AI Security Review + pentest of the API and web application around the model (ASVS L1/L2); one report, free retest | approx. 15 working days |
| AI Evals in CI | a suite of prompt regression tests (promptfoo) and chat interface e2e tests (Playwright) wired into GitHub Actions / GitLab CI; handover to the team | approx. 5 working days |
| Web / API Pentest | classic pentest without the AI layer, per OWASP ASVS, report, free retest | approx. 10 working days |
| Performance Testing | k6 scenarios (including LLM endpoints), execution, report with thresholds | approx. 5 working days |
| Retainer | 4 working days a month: AI testing at major prompt or model changes (evals in CI at every change), test automation, consulting; priority in the queue | monthly |
Everything that is yours: the system prompt, the knowledge base (RAG), the tools the agent can access, permissions, input and output filtering, limits, logging. That is where most vulnerabilities arise, not in the model itself.
Yes. The test checks what the bot can reveal and do in your context (data, integrations) and whether it meets the AI Act transparency obligations; you also get a list of questions/fixes for the vendor. Some risks sit with the vendor, and the report separates them. If the bot runs on the vendor’s infrastructure, the scope takes the vendor’s security testing policy into account and, where needed, its consent.
We agree on a token budget in the scope. Testing runs on a test environment or with limits. There are no load tests on a production key without consent.
Each project is quoted individually after a 30-minute call. You get a quote with the price within 2 working days, before anything starts. The price depends on the application size, the number of models, integrations and agent tools, and access to a test environment and documentation. A retest after fixes made within 30 days of the report is included.
AI Security Quick Check: 5 working days. Full AI Security Review: approx. 10 working days; with an API pentest: approx. 15. We set the start date on the call.
A PDF report (PL or EN): an executive summary, a list of findings with priority, example prompts/requests and recommendations, scope and methodology. Plus a walkthrough meeting and a retest.
It confirms that the application went through testing and a retest within the scope described in the report, in the stated month. It does not mean the application has no vulnerabilities.
Testing runs on the environment you designate. On production, only non-invasive tests run by default, in the agreed window; load or invasive tests only with your explicit consent. Agents with access to real systems (e-mail, payments) are tested exclusively on test accounts.
Yes, by default, before discussing details, on your template or on the codequality.pl one.
No. It is technical evidence an auditor usually wants to see, plus help preparing that part; the formal audit is done by an auditor or a law firm.
Yes, the company is VAT-registered in Poland.
A few sentences are enough: what your application does, where it uses AI and what you want tested. The reply comes within 1 working day, with a proposed time for a free call.