Skip to content
Research

Notesfroman
autonomoussecuritylab.

Field reports, technical notes, benchmarks, and papers from the CYBRET research team. Published when something is true, not on a cadence.

CITATION GRAPH · LIVE
37 NODES · 94 EDGES
r·001r·002r·003r·004r·005r·006r·007r·008r·009r·010r·011r·012r·013r·014r·015
35
Notes & papers since 2024
2,271
Labeled graphs in CYBRET-Bench
39
Agentic-attacker capabilities catalogued
FILTER /
39 RESULTS
FEATURED PAPER, Q1 2026

The CYBRET Thesis: Defense at Machine Speed.

68 pages on neurosymbolic reasoning, adversary modeling, and the economics of agentic defense. Includes the full CYBRET-Bench v0.3 protocol and reproducibility scripts.

thesis.tex · abstract
\section{Abstract}
We propose a neurosymbolic framework for reasoning over heterogeneous trust graphs, in which autonomous agents reconstruct exploitable paths across code, cloud, identity, and runtime signal layers. On CYBRET-Bench v0.3, our system recovers the ground-truth attack path with 0.914 reachability-precision, under a 400 ms end-to-end latency budget.
% 68 pp · 41 figures · MIT license
Subscribe

A short note when we publish. Nothing else.

Five or six times a year, one paragraph, a link, signed. No drip campaigns, no product announcements, no webinar invites.

SIGNED · DKIM · NO TRACKERS