Observability is the next generation of endpoint security. SACR just mapped the market.
The agents you can't see
For decades, the risk on your endpoints came from outside. Someone broke in, or something malicious slipped through, and your defenses were built to catch it. The new risk is getting invited in. Companies are handing AI agents the keys and asking them to be as productive as possible, and almost nobody can answer two questions about them. Where are your agents, and what are they doing?
A new analyst report maps that gap. On July 22, Software Analyst Cyber Research (SACR) published its report on Endpoint Control and Prevention, a marketing map of the vendors securing the layer where AI agents now do real work. Origin sits in Zone 3, agent runtime observability, the zone that answers what your agents actually did. We are firm believers that endpoint security is heading toward observability, and the report lands in the same place.
How the endpoint got here
Every endpoint tool was built to chase a new threat. The first was file-based malware, and antivirus answered it by matching every file against a database of known signatures. That held while malware was a static file. Then attacks went fileless. Adversaries stopped dropping binaries and started working through the machine’s own processes and tools, so EDR answered by watching process behavior for attacker behavior. Each generation matched the threat of its era, and both still do their jobs today.
The report argues that the threat is now moving up the stack again, past the file and past the process, to the agents working on the endpoint. This third vector is non-deterministic. An agent arrives invited, holding real credentials. It works at machine speed, and it will not do the same job the same way twice, so there is no signature for what it does next. It can do damage while looking exactly like the work you hired it for. Nothing built for the first two vectors can tell the difference.
This is the gap SACR’s Endpoint Control and Prevention framework was built to cover, five zones running from software posture through data controls. Zone 3 covers runtime observability, reconstructing how a prompt became a tool call, then an execution step, then a response from the system. That is what Origin captures. A user-mode sensor on the endpoint records the full local trace of every agent action, at the layer above the operating system where a prompt turns into work. Your EDR keeps watching the process. Origin watches the reasoning layer stacked on top of it, and the two run side by side.
Why observability is security
Novel AI-era attacks are anomalies of intent. A dangerous session and a productive one can run the same model and the same binaries. Both open a shell, edit a file, call an API, reach out over the network. What separates them is why those actions happened and how they relate to one another.
A signature cannot see intent. A process tree cannot reconstruct how the actions relate. You need the trace.
Origin baselines the normal relationship between prompts, tools, and the actions they trigger downstream. Then it surfaces the moment behavior departs from that baseline, even when every individual action looks legitimate on its own. Reconstruct what happened and you can separate useful automation from a real threat. Without that reconstruction, you are guessing. On the agentic endpoint, observability is security.
What that looks like in practice
Start with an investigation. An agent did something overnight, and someone wants to know what. Ask in plain language what the agent did, and the record comes back with the evidence attached, the prompt that opened the session, the tools it called, the files it touched, the data it moved, and where it failed. You start from a connected narrative of what happened and why, not a pile of event logs to stitch together by hand.
When something looks wrong, replay the work and watch it happen. A build step that quietly printed a cloud credential into its output shows up in the trace the moment it runs, days before anyone would have caught it, if they ever did at all.
The same trace shows every agent on the fleet, including the ones nobody approved. And because you own the record, the questions do not have to stop at security. Ask what your AI spend produced last week. Token bills tell you what you spent. Origin tells you where your intelligence actually went inside the organization.
See it on your own endpoints
The report gives this shift a name and a market map. Without endpoint observability, you can say what you licensed your agent workforce to do, not what it actually did. That is the gap Origin closes.