Level: Tehnical

Abstract:
We spent two years worrying about LLMs hallucinating. Now we have to worry about them having “hands.”

Coding agents are evolving from passive autocomplete into autonomous engineers that run commands and commit code. This changes the threat model. The attack surface has expanded from a chat window to your local machine, credentials, and production servers.

This talk covers how the autonomy that makes agents useful also makes them vulnerable. I will walk through the progression from early IDE bugs to attacks that use an agent’s helpfulness against it. Traditional security boundaries fail when an attacker can manipulate your trusted tools.

If AI can act, it can be weaponized.

Bio:
Aleš Brelih – Security engineer @ 3fs – Personal website: https://alesbrelih.dev/about/

Posted on Tuesday, March 10th, 2026 at 3:02 pm in talks | rss feed for comments| Both comments and pings are currently closed.