What we are
SGAIL Labs is a Montana LLC and Delaware C-Corp based on the North Shore of Oʻahu, Hawaiʻi. We build AI security infrastructure across three lines of work: the SGAIL Firewall and its supporting stack, an adversarial-input detection family of open-source tools, and research into new architectures, datasets, and training methods.
How we build
The way we build is the product. A few principles run through everything we ship:
Minimal surface area
Conservative by design. Fewer dependencies, smaller attack surface, offline-first where it counts.
Compartmentalization
Components and agents stay in their lanes. We favor compartmentalization over consolidation.
Evidence woven in
The witness layer is not an afterthought. Every consequential action leaves a tamper-evident record.
Hardware-first
Experience-grounded and proven over clever. Simplicity is a signal of correctness.
The SGAIL Firewall
Our flagship line is a deployable control point for AI-to-AI systems. Its public core is the Harborlight Firewall — a tamper-evident, Merkle-chained witness and enforcement layer — extended by the adversarial-input detection family that inspects and grades traffic before it reaches a model. It runs offline, holds risky exchanges pending review, and preserves a defensible chain of custody.
Training data & methods
Because our detectors are real, the data they produce is real. We package the adversarial samples, red-team corpora, and evaluation methods that come out of our security work and make them available to other teams training models that need to survive hostile input.
Where we're headed
We are pursuing DHS SBIR funding to accelerate the security-infrastructure work, while continuing to publish open-source detection tooling and take on dataset and research engagements. If any of that overlaps with what you're building, we'd like to hear from you.