About Liana Tumescu

Liana Anca Tomescu is the founder of startup SonnyLabs.ai, which ensures AI apps don’t get hacked and stay secure. She launched a podcastabout AI Security called “AI Hacks with Liana”, found on Spotify and YouTube. She has a background in cybersecurity, having previously worked at Microsoft and Workday, and having completed a masters in cybersecurity from UCD.

SonnyLabs,ai is building the essential security layer for agentic AI. Our platform detects and blocks prompt injections, tool misuse, and data exfiltration in real time across AI agent workflows. We help regulated organisations ship AI agents faster, clear security and compliance reviews, and turn secure AI into a measurable competitive advantage. Schedule a call with Liana to hear how to secure your AI agents here.

Your AI Is Being Hacked With Sentences, Not Code

As the age of AI moves from chatbots to agentic systems, we are seeing a new category of
security failure: systems that follow instructions they should refuse, access data they were
never meant to see, and trigger actions no one approved, all because of carefully crafted
language. Security problems caused not by malware or by broken encryption. In her talk, Liana will break down how several threats, including prompt injection, agent manipulation, and tool abuse impact AI apps and AI.