Strahl — Trust at machine speed

Humans of Strahl

Strahl is an applied AI research and product lab. We are building a future where AI is dependable enough to become usable at scale for the work that matters to humans, and to make its risks measurable, manageable, and transferable.

AI agents are beginning to act on our behalf: handling sensitive information, making decisions, and coordinating with other systems. As their speed and autonomy increase, human oversight cannot remain the foundation of trust. We believe safety should be a property of how a system is built. An agent’s access to information, authority to act, and obligations to others should have precise definitions and enforceable boundaries.

We are a team of researchers and engineers dedicated to studying how agents use information, how to constrain their behavior, and how to establish that those constraints hold.

Good brakes make cars faster

Historically scientific progress has required practical safety innovations to diffuse into our societies. Elevator emergency brakes enabled the high rise boom, fuzes were pivotal in bringing electricity into our homes, and more recent innovations like SSL have reshaped internet commerce.

We see this pattern repeating today. AI is already incredibly capable, and delivering immense value to the early adopters that can structurally stomach more risk than the rest of society. Coding tools are reshaping how tech companies build and conduct research, compressing timelines and increasing competitiveness. Meanwhile, the bulk of value-creating workflows are locked behind much stricter needs of fundamental system safety. Without guarantees on behavior, regular enterprises can not adopt these technologies for sensitive work. At Strahl, we are building the practical safety infrastructure that will enable safe deployment of AI based systems at a previously unprecedented scale.

Fundamental safety

Beyond practical safety, we are also looking at advancing intrinsic human safety of AI systems as a whole. We are particularly interested in analyzing how multi-agent and distributed AI systems interact with each other and with humans. If we can secure a single agent, then we can compose this onto arbitrarily many agents.

Join us

We are pushing the frontier of safety and verification for AI agents to enable AI systems that are safe to deploy at scale. We value curious researchers, who love hard engineering problems, and connecting the dots across multiple domains to solve the hardest problems in the most creative and practical ways.