Cybersecurity

Stihia.ai: how a Bulgarian startup protects companies from AI threats

Stihia.ai started with a product for protecting AI agents and pivoted to training and AI policies. The team, its first clients and the lessons your company can take from it.

Stihia.ai: how a Bulgarian startup protects companies from AI threats

The more companies roll out AI agents and assistants, the more pressing the question of how to protect them becomes. Stihia.ai is a Bulgarian startup working on exactly that question. Its story is also useful because it shows how a product changes once it meets the real market.

The team

  • Ivan Ivanov and Petar Kirkov are responsible for technology, machine learning and cybersecurity. Kirkov served as Bulgaria's national cybersecurity coordinator until 2024.
  • Stiliyan Mitrev is responsible for the product vision.
  • Stoyan Simov leads business development and sales.

The company was registered in August 2025 under the name "Stihia". It grew out of a game by Ivanov, Stihia Zmey, in which players try to extract protected data from an AI system. The game later became part of the company's services.

From product to consulting and back

The first product, Stihia Sense, was built for real-time threat detection at companies running their own AI agents: blocking prompt injection, data leaks and agents that go out of control. Attempts to sell it to banks and fintech companies, however, were slow to take off.

The team pivoted to training and consulting, including internal AI policies aligned with the AI Act, GDPR and DORA. The first client was the tech company Dreamix, in June 2026. The focus is on Bulgarian software companies with 30 to 70 people, and there is also interest from public institutions such as the Bulgarian Academy of Sciences. The company has so far been self-funded and is open to strategic investors.

The next step

Stihia Factory, a working name for now, is meant to build security into the development process itself, with a focus on the risks of AI coding assistants.

What other companies can take from this story

  1. An AI policy is a practical document. Which tools are used, what data may be entered into them and who approves it.
  2. Code from AI assistants gets reviewed. Especially when it touches security, payments or personal data.
  3. Training is part of your defense. People need to recognize manipulation aimed at AI systems.

In our web development work, we use AI tools carefully and every piece of code goes through review. Related: AI agents in the role of cybercriminals.

Topic source: Digitalk: Stihia.ai, the Bulgarian answer to the AI cyber threat (in Bulgarian). This article is an independent analysis by Penkov Studio.

View all
Ransomware Attacks Target Managers: What the Data Shows

Cybersecurity2 min read

Ransomware Attacks Target Managers: What the Data Shows

An analysis of more than 350 incidents shows the typical ransomware victim is a manager in their mid-forties. What that means for protecting your company's data and website.

Next step

Got a project in mind? Let us turn it into a website that performs.

Tell us briefly what you need. You will get an email reply with a few clarifying questions and an indicative quote.