Mostly
Harmless

An AI lab building, auditing, and teaching the systems that build the systems — quietly, and with a towel.

What we do

We build, audit, and teach AI systems that act on their own.

01AI Security

We find what your model didn't expect to be asked. Adversarial testing, threat models, and audits for production AI.

02Training

Specialized training for technical teams putting AI into production. Practical, deep, and free of magical thinking.

03Agentic AI

Designing and shipping autonomous systems that mostly behave themselves. From prototype to production.

04Research

Public notes from the strange edges of the field — safety, alignment, and the experiments worth telling you about.

Get in touch

info@mhl42.ai

We read everything. We answer most things.