Ghostfield researches mathematical foundations for AI safety in safety-critical and mission-critical applications. We investigate how AI can build and operate software within verifiable constraints, and how those constraints can be built into intelligent systems themselves.
Our systems engineering focuses on bare-metal and embedded Rust, deterministic behavior, and software assurance.
Four connected areas supporting AI safety research, grounded in mathematical foundations and the behavior of real systems.
01
AI-assisted software development
AI-generated software changes need to preserve the rules of the systems they affect. We research how explicit specifications and independently checkable evidence can constrain those changes, including their effects on ownership, composition, and failure behavior.
02
Constrained AI operation
An AI system's actions can change software, consume resources, and produce real effects. We investigate how permitted actions, resource bounds, and failure behavior can be made explicit and independently checkable throughout operation.
03
Mathematical foundations of intelligent systems
Reasoning, memory, learning, and action must remain governed as an intelligent system changes. We research how to construct intelligent systems with constraints built into their structure, and how those constraints behave as components interact.
04
Bare-metal Rust and software assurance
Software for safety-critical and mission-critical applications must account for hardware, memory ownership, timing, and resource limits. Our systems engineering focuses on bare-metal and embedded Rust, deterministic behavior, and evidence that connects mathematical claims to implemented software.
02 / Research
Mathematical foundations for AI safety
A constraint that holds for one component or operation may not survive its interactions with the rest of a system. We investigate how mathematical foundations connect local behavior to system-level properties across state, resources, composition, failure, and change.
How can intelligent systems preserve their constraints as they reason, learn, change software, and act?
Research direction 01
Safe AI development and operation
We research how the software AI creates and the actions it takes can remain within explicit, independently checkable system constraints. The aim is to govern changes and effects throughout development and operation.
Research direction 02
Mathematical foundations for intelligent systems
We investigate how to construct intelligent systems whose reasoning, memory, learning, and actions remain governed as they change. The aim is to build constraints into the systems themselves.
Evidence before confidence
Our approach makes the conditions of a claim part of the work.
I.
Define the permitted behavior
Make assumptions, constraints, and observable consequences explicit.
II.
Connect parts to the whole
Examine how interactions affect the properties of the complete system.
III.
Account for change and failure
Treat resource limits, partial effects, and subsequent behavior as part of the problem.
03 / About
Washington State, United States
Ghostfield LLC
Ghostfield is an independent research and systems engineering company based in Washington State. We research mathematical foundations for AI safety in safety-critical and mission-critical applications, connecting intelligent-systems research with the practical limits of software, hardware, and physical systems.
Focus
AI safety and mathematical foundations for intelligent systems
The person behind the work
Aaron Watts
Founder Research & systems architecture
Aaron Watts is the founder of Ghostfield, where he leads research and engineering in software for safety-critical and mission-critical applications. Previously, he led machine learning initiatives in the public sector, taking technology programs from initial concept through systems design, policy, contracting, and implementation planning.
We welcome research collaborations in AI safety and intelligent systems for safety-critical and mission-critical applications, including embedded Rust and software assurance.