Case study · Axon Labs research · Glasgow partnership
AI cybersecurity at the memory layer
Applied research into intelligent protection closer to hardware and memory — models and signals software-level attackers struggle to fake.
Software agents can be lied to. Hardware signals are harder to fake.
This research line explores whether compact models, run close to memory and bus behavior, can surface compromise that traditional endpoint tools miss — and whether that detection can be productized into real devices.
Program outcomes
- Research directed at detection signals below the OS trust boundary.
- Model and architecture choices constrained by continuous on-device budgets.
- Public scientific and IP claims gated on academic and legal approval.
Research thesis
Intelligence that survives compromise above it.
If the OS is owned, many security agents are owned with it. Ground-truth signals from lower layers — access patterns, timing, power signatures — create a different detection surface. The open problem is doing that continuously within product silicon budgets.
Partnership
University of Glasgow · PhD-level collaboration
Axon Labs participates in a formal research partnership focused on AI cybersecurity. Detailed claims, publications, and IP language are published only with academic and legal clearance.
Product path
From research artifact to architecture option.
We treat research as a path to owned capability: model families, interface requirements, and secure response paths that can later appear in client product architectures — not as a poster on a conference wall alone.
Need this level of engineering ownership on your program?
Tell us the system, the constraints, and the decision you need next. We will tell you honestly whether Axon Labs is the right engineering partner — and at which phase to start.