The most capable AI ever built is sitting on the sidelines. Not because it can't do the work, but because no one can yet trust it to act where the stakes are real: money, medicine, the decisions that shape people's lives.
We're building what changes that. When an AI can be trusted to act, and answer fully for what it does, it can finally be let into the places that need it most. This is the work that decides how far AI is allowed to go, and there is nothing more important for us to be doing.
We're a team in San Francisco and Los Angeles, building it now. If you want to do the best work of your career on a problem that matters, come build it with us.
Explore open roles{{ b.lead }}
{{ b.body }}
{{ b.invite }}
Every action an AI takes passes through MOSS first. The dangerous ones don't make it through.
An agent reaches for data it shouldn't touch, and the call is blocked before it lands. One goes off the rails, and a kill switch halts it mid-run. Another starts to drift off course, and MOSS flags it early.
The hard part is doing this in the hot path. Policy has to be evaluated inline, in single-digit milliseconds, so the agent never feels it. Every call it does make carries an ML-DSA-44 signature and a tamper-evident record, at the rate a fleet of agents generates them. And the tool calls run over MCP, which has no signing standard yet, so you design the manifest format as you build it.
Don't see your role? If you're one of the best in the world at something we need, write to us anyway.
Tell us about something hard you've actually done: a paper, a working group you chaired, a deal you closed, a system you kept alive at 3am. That's what we read first.
Skip the cover letter. Conversations are informal and technical, and you'll have as much time to interview us as we have to interview you.
careers@mosscomputing.comMOSS is an equal opportunity employer.