The enterprise AI landscape is undergoing a fundamental shift. We are at a technological inflection point where Large Language Models (LLMs) and their downstream implementations as autonomous AI agents are moving from pilot programs into core enterprise infrastructure, running projects end-to-end with minimal human oversight. Industry projections suggest that nearly half of all generative AI adopters will deploy agentic applications within the next two years.

At the same time, this transformation has opened one of the most serious and urgent security challenges in years. AI agents introduce entirely new ways for attackers to cause harm, and existing security tools were not designed to address them. Traditional cybersecurity relies on fixed rules and predictable behavior, which works well for detecting malicious code or obvious errors. But attacks on agentic AI systems exploit the fact that these systems think in probabilities and can behave unpredictably. 

As AI agents become widespread, they introduce new risks, such as being tricked into following harmful instructions (prompt injection), unintentionally exposing sensitive information, or being granted excessive autonomy that can lead to severe unintended consequences. Because LLMs ingest and process vast amounts of internal company data, AI agents can inadvertently leak confidential customer information, proprietary business strategies, or source code through seemingly harmless interactions. Furthermore, when agents are empowered to execute actions, such as processing financial transactions or modifying critical system configurations, a single error or malicious prompt can cause catastrophic enterprise-wide damage, often without human oversight.

Enter Pallma AI, a company building the security layer for autonomous AI agents. Pavlos Mitsoulis and Dionysis Varelas are AI experts, each with over 13 years of experience building machine learning and large-scale engineering systems at companies including Expedia and Activision Blizzard (now Microsoft Gaming). They have been training and deploying large-scale AI models since 2016. A few months ago, they started Pallma AI to secure agentic enterprise systems, enabling businesses to safely unlock AI’s productivity potential.

Pallma AI is the first line of defense for this new world: an AI-native security platform purpose-built to protect AI-powered applications by detecting threats in real time and continuously improving agent security. The platform integrates seamlessly into existing enterprise technology stacks, collecting application and system data and using advanced AI models to monitor, detect, and neutralize threats targeting AI agents in real time.

Pallma AI is designed to function as the central nervous system for AI security through real-time threat intelligence (blue teaming) and proactive defense and autonomous hardening (red teaming). It uses AI to summarize complex threats, provide clear recommendations, and offer the option for automated resolution. At the same time, they go beyond passive defense. They develop advanced AI models to identify vulnerabilities before malicious actors can exploit them and recommend necessary fixes, ensuring AI applications remain continuously hardened against emerging threats.

We are excited to back Pavlos (CEO), Dionysis (CTO), and their team on their journey to help organizations detect, prevent, and manage risks in the era of agentic AI. Today, the company announces the completion of a USD 1.6 million Pre-Seed round led by Marathon Venture Capital with participation from technology leaders from AWS, Meta, Google, Hack The Box, Workable, and others. The new funding will enable Pallma AI to expand its team across London and Athens, accelerate an ambitious product roadmap, and meet rapidly growing demand for its AI-powered security platform.

The rapid adoption of AI agents represents a new frontier in cybersecurity risk. These systems have deep access to enterprise infrastructure and are becoming prime targets for malicious actors. This is not theoretical; it’s already happening. The combination of asymmetric risk and increasing regulatory pressure around AI governance has created a significant market opportunity. We believe Pallma AI is uniquely positioned and entering the market at exactly the right time.

Building with AI agents? Request access to Pallma AI here.

Pallma AI is an AI-native security platform purpose-built to protect agentic enterprise applications by detecting threats in real time and continuously improving agent security

Learn more