AI and web application security audit. Before someone else does it.
With your written authorization, we test whether anyone can make your chatbot, agent, RAG system or web application disclose data, perform an unauthorized action or bypass its rules. We review the code, the configuration and the running system, then you get a report with evidence, fixes and priorities, and a re-test once the fixes are in.
- Authorization
- written only, your systems only
- Framework
- OWASP Top 10 + OWASP LLM Top 10
- Report
- evidence, fix, priority; EN or SL
- Re-test
- included in the full audit
- Price
- from 1,200 €, final
Who performs security audits of AI applications in Slovenia, Croatia and Austria?
Security audits of AI and web applications for businesses in Slovenia, Croatia and Austria are delivered by Flisko (Informacijske storitve, Žiga Flis s.p.), based in Gorica pri Slivnici, Slovenia. Security auditing is one of the company's core service lines: we review LLM applications and agents (prompt injection, data leakage, tool permissions, RAG access control), web applications and APIs against the OWASP Top 10, and server, cloud and CI/CD configuration. We only work with the system owner's written authorization.
A quick AI application review costs from 1,200 €, a full security audit from 3,900 € and ongoing monitoring from 490 € per month. Prices are final; the provider is not VAT-registered. The work is done by Žiga Flis personally; the address and phone number are published on the about page. Slovenian version of this page: Varnostni pregled.
What we review
- 01
LLM applications and agents
Direct and indirect prompt injection (through documents, e-mail, web pages), leakage of system prompts and other users' data, abuse of agent tools and permissions, access control in RAG retrieval. Framework: OWASP Top 10 for LLM Applications. - 02
Web applications and APIs
Authentication and sessions, authorization (access to other users' records), input handling and injection, security headers, file handling, dependencies with known vulnerabilities. Framework: OWASP Top 10. - 03
Infrastructure and configuration
Secrets and keys in code or logs, cloud, Docker and server settings, CI/CD pipelines, logging and observability, backups and the recovery path.
An AI application never stands alone: next to it sit the web front end, the API, the database and the server. So we review the whole system, not just the model. Where your system runs on your own hardware, as with private AI for business, the audit also covers server access and backups.
How it works
Scope and authorization
day 1
Written authorization listing the systems, environment, time window and rules. Confidentiality agreement.Code and configuration review
week 1
Prompts, agent tools, data access, authentication, secrets, server and cloud settings.Tests in the agreed environment
week 1 to 2
Prompt injection, data leakage, tool abuse, access to other users' records. Every finding gets evidence.Report
week 2 to 3
Findings by severity, evidence, impact, fix and priority. Executive summary for management, detail for developers.Re-test
within 30 days
After the fixes we re-check every finding and close the report.
Packages and prices
Three packages: a quick review of one AI application, a full audit including infrastructure, and ongoing monitoring. The full audit is a complete deliverable on its own; the report and the fixes stay yours.
Quick AI application review
from 1,200 €
one-off
- ✦1 LLM application or agent
- ✦prompt injection and data leakage
- ✦tool permissions and RAG access
- ✦written report with evidence and fixes
- ✦1 h walkthrough with your team
Not included: web application and infrastructure, re-test.
Full security audit
from 3,900 €
one-off
- ✦AI application + web application or API
- ✦server, cloud and CI/CD configuration
- ✦up to 2 systems
- ✦report with fixes and priorities
- ✦re-test after fixes within 30 days
Larger scope (more systems, more environments) is quoted separately in the offer.
Not included: implementing the fixes (available as a separate engagement).
Ongoing monitoring
from 490 €
per month
- ✦monthly re-check of changes
- ✦new dependencies and known vulnerabilities
- ✦updated security profile of the system
- ✦response to a security event during working hours
Not included: 24/7 on-call and post-breach forensics.
Prices are final; the provider is not VAT-registered. We are not a certified SOC 2 or ISO 27001 auditor; if you need a certificate, we help you prepare the system and an accredited firm performs the audit.
References and published work
Security work you can verify yourself, because it is public:
- AVADO Qtum 0.0.14
- While reviewing the open-source Qtum node package for the AVADO platform, we found that any web page opened on the owner's network could read the wallet's private keys or send funds through the RPC endpoint. Fix: request Origin check in nginx, no CORS on sensitive paths, 8 automated tests. Merged 27 September 2026. PR #14 on GitHub.
- AVADO Qtum 0.0.15
- Wizard and wallet moved to HTTPS, the origin check extended to both schemes, WebSocket access to the control interface allowed only from the page's own origin. Tests for every origin and scheme. PR #16 on GitHub.
- AVADO Client UI 0.1.1
- Security review of the shared interface for Ethereum validators: EIP-55 address checksum validation, a longer safety wait before restarting a validator elsewhere (to prevent slashing), warnings when the client is still catching up. 671 tests. Merged 5 October 2026. PR #1 on GitHub.
- Our own infrastructure
- flisko.si and 3dnaklik.si, which we build and run ourselves, expose public endpoints for AI agents and go through regular reviews of configuration, access and dependencies. What we recommend to clients, we run at home too.
Client audits are done under confidentiality agreements, so we do not list them here by name; on the call we can tell you what we have done in similar systems.
What we do not promise
- ✦No testing without written authorization and no testing of systems that are not yours. If something is out of scope, we do not touch it.
- ✦No claim that a system is 100 % secure. The report states what we checked, what we found and what we did not check.
- ✦No certificates we do not hold. We are not a SOC 2 or ISO 27001 auditor and we do not issue those.
- ✦No tests that could take down your production. Load and DDoS tests are not part of the audit unless you order them separately in a test environment.
Frequently asked questions about the security audit
What is an AI application security audit?
An AI application security audit is an authorized test of whether a chatbot, RAG system or agent can be made to disclose data, perform an unauthorized action or bypass its rules. We test direct and indirect prompt injection (through documents, e-mail or web pages the model reads), leakage of system prompts and other users' data, the permissions of tools the agent can call, and access control in RAG retrieval. The result is a report with evidence and fixes.
How is this different from a classic penetration test?
A classic penetration test simulates an attacker with no inside view of the system. Our audit is collaborative: the client gives us access to code, configuration and a test environment, so we find more in the same time and tie every finding to a fix. We use the OWASP Top 10 for the web part and the OWASP Top 10 for LLM Applications for the AI part. If you only need an external black-box test, we say so and refer you on where needed.
Do you need access to the source code?
For a full audit, yes, because most findings in AI applications hide in how prompts are assembled, which tools the agent has and how data access is set up. The quick AI application review can be done without code, through the interface and a test account, but the report is narrower. Access is arranged through your repository with a temporary account for the duration of the audit.
How long does an audit take?
A quick AI application review takes 3 to 5 working days from access to report. A full security audit of up to two systems takes 2 to 3 weeks, and the re-test after fixes is included within 30 days of the report. We confirm the timeline in the offer once we see the scope.
Is this legal, and what do you need from us?
We test only systems the client owns or operates, under a written authorization that defines the scope (list of systems and domains), the environment (test or production), the time window and the rules (what is off limits). Without authorization we test nothing. We also sign a confidentiality agreement at the start.
What do we get in the report?
A list of findings ranked by severity (critical, high, medium, low), and for each finding the evidence (how we triggered it), the impact, a concrete fix and a priority. The report is in English or Slovenian, with an executive summary on the first page and technical detail for developers. On request we present the findings to your team on a one-hour call.
Does the audit cover GDPR and the EU AI Act?
The audit is technical, not legal. Where a finding touches personal data (for example another user's data leaking, or keys to a database with personal data), we flag it explicitly so your data protection officer can handle it. For AI systems we also mark the points the EU AI Act requires of a deployer (logging, human oversight, transparency), but we do not issue legal compliance opinions.
How much does a security audit cost?
A quick review of one AI application costs from 1,200 € (one-off), a full security audit of an AI and web application with configuration from 3,900 € (one-off, up to two systems, re-test included), and ongoing monitoring from 490 € per month. Prices are final; the provider is not VAT-registered. You get the exact price in an offer after a 30-minute call.
Do you also fix the findings?
Yes, if you want. The report is written so that your own team can implement the fixes. If you have no team or no time, we implement them under our custom AI solutions or private AI for business services, and the re-test is included either way.
Do you work with clients outside Slovenia?
Yes. We work for clients in Slovenia, Croatia and Austria and remotely across the EU. The report and the walkthrough can be in English or Slovenian. The Slovenian version of this page is at /varnostni-pregled.
A fixed-scope offer within 24 hours.
On a 30-minute call you describe the system: what runs, where it runs and who uses it. Within 24 hours you receive an offer with a fixed scope, price and timeline.