#UYBHYS26

Your cloud environment has privilege escalation paths you haven't found yet. Static tools dump tables of misconfigurations, but they don't show you how those misconfigurations chain together into full attack paths or which ones an attacker would actually find first. Security teams need to see the shape of their exposure, rather than searching through spreadsheets of individual findings stitched together by guesswork.
PathQuarry is a visualization engine that lets you watch a reinforcement learning agent explore your cloud attack surface in real time. The entire tool is open source and extends to Azure and GCP. Feed it an AWS environment, and a Datalog engine derives what's reachable like IAM misconfigurations, overpermissioned roles, cross-service trust relationships, and resource-level policy gaps, while a Q-learning agent discovers and ranks multi-step escalation chains live on screen, rendered as an interactive force-directed graph. Nodes light up as the agent visits them. Edges thicken as paths get reinforced. Dead ends fade out. You can pause, rewind, step through the agent's decisions at each state, and see exactly why it chose one path over another.
I will demo PathQuarry against AWSGoat and CloudGoat, two deliberately vulnerable AWS environments, showing how the visual approach surfaces non-obvious attack chains: lateral movements through Lambda execution roles, cross-account pivots via confused deputy patterns, and credential chaining through instance metadata that get buried in text-based output. I will also show how the agent's learned Q-values provide a natural risk ranking: the paths it converges on fastest are the ones a real attacker would discover first. By visually proving this convergence, we give an opportunity for defenders to prioritize remediation in a principled way.