Task completion is an incomplete measure of AI work. Agent traces let us infer an organization’s norms and intents, then evaluate behavior against its purpose, practices, and policies. Securing AI work requires modeling the organization. originhq.com/blog/norms-and-…

Defense has fundamentally shifted away from an external actor/campaign-centric model toward insider threat, but so much of what we’ve built over the past 15+ years wasn’t designed for this. Years of ATT&CK bingo have atrophied the muscles required to combat subtle acts of subversion, and now they’re occurring at machine speed and often without the users’ knowledge. Countless teams I talk with are drowning in alerts generated by agents operating in massive grey areas, doing things that look malicious but are completely benign and vice versa. The missing piece is, as it always has been, the intent behind the action. False positive tuning was already the painful exercise of explaining that intent in org-specific context so we could make exceptions for acceptable behavior. Agents make that substantially harder through their speed, volume, and contextual ambiguity. The next generation of security controls has to reason about intent in organizational context, because the behaviors we’ve spent many years learning to detect are becoming indistinguishable from the work we’re paying agents to do.

Our team has been testing this vector in the @originhq platform this morning and we're seeing strong detections fire via auto trace review. If you're interested in this technique, @depletionmode put out a blog a few months ago on a similar area: originhq.com/research/the-mo…

I'm hiring Forward Deployed Engineers at @originhq. The work is at the frontier of observability and AI, embedded inside the most important, most forward-leaning organizations in the world. Must be in the US or Canada. DMs open. jobs.ashbyhq.com/originhq/aa…

Claude being flaky on a hard deadline is a good reminder that single-provider agent stacks are a bad long-term bet. I don’t want my workflow to depend on one lab being up, fast, cheap, permissive, and best-in-class forever. The winning agent environments will be horizontal across providers.

Phenomenal work from @tyholms

The endpoint telemetry we've spent a decade pointing at adversary detection can answer a much broader, and increasingly more interesting, set of questions. You can understand the types of problems people are solving, how they approach their work, where intelligence is being deployed and how people are extracting value from agents, and what types of work are being duplicated (or more) because we lack the ability to capture output. These are massive questions and most companies are guessing at all of them. The same telemetry that flags a credential dumper can also tell you which team has cracked a workflow worth handing to the rest of the org. Security is only a fraction of the value. We're building @originhq to turn that data into all of those answers. Once you can see what people and their agents are doing on an endpoint, you can answer almost any question worth asking about how your company actually runs. AI is making the set of those questions wider, and it's making the answers cheaper to produce at the same time. This is the work I left red teaming for.