Lobster Labs is the umbrella for products, experiments, and research systems spanning AI safety, alignment & compliance, security, legal, and physics. We ship real tools — drivable by humans and the agents acting on their behalf.
We build AI that reduces your need for AI.
Not a contradiction — the whole point. The danger isn't that AI is wrong; it's that it's always available. We design tools that clear bureaucratic fog and hand judgment back to you, with friction where deference would otherwise creep in.
Products / shipping today
Take the reviews you wrote on Google Maps, sign them with Ed25519 and did:web, and sell agents access over x402 micropayments. Three priced tiers hand back different artifacts — readable text, structured fields for inference, or the full signed credential for training — so the price maps to a stated licence rather than a label.
Runtime governance for AI agent tool use. Evaluates an attempted action in context and routes it to allow, review, or block — with an audit trail. Built for the gap static permissions leave behind: same tool, different context, different risk.
A personalized, reversible disk-cleanup CLI. 91 declarative rules, a 7-day airlock, and pressure-tests before anything is touched. One binary. No GUI. No cloud. No telemetry. No sudo.
Enforcement infrastructure for AI-era content licensing, provenance, and active defense — the layer between your content and the AI systems that want to train on it.
Bring-your-own email infrastructure for agents, workflows, and programmable communications. Full API control, your domain, your mailbox.
Research / frontier
A concept paper on proton–boron propulsion, direct collimation of nonthermal fusion exhaust, and a near-term research-prioritization argument for getting there. The lab's frontier track — where the physics is as serious as the software.
Read the p-B11 paper →Track record / shipped