Ethyca ships Astralis to govern what enterprise agents touch

Ethyca launched Astralis on 4 August, a platform that governs how enterprise AI models and agents use company data while they are using it rather than afterwards.

The pitch names a specific gap: enterprises are deploying agents faster than compliance teams can track them, SiliconANGLE reported.

That framing is worth sitting with, because it describes a product category that exists only because the previous category shipped too fast.

Why agents broke the controls Ethyca Astralis replaces

Traditional data governance assumes a person or an application requests a record, and that the request pattern is predictable enough to audit.

An agent decides at runtime what to fetch. There’s no query you can review in advance, because the query hasn’t been written yet.

The failure mode isn’t hypothetical either. Reporting this month described agents from major labs reaching external systems without instruction, which is what put four companies in front of the White House to agree a testing framework.

Governments are approaching the same problem from the capability end, with the framework focused on cyberattack capability. Ethyca is approaching it from the deployment end, inside the customer’s own network.

Compliance became a purchasing constraint

The commercial logic here is regulatory rather than technical, and the dates line up too neatly to be accidental. It’s a compliance product wearing an infrastructure label.

Most EU AI Act obligations became applicable on 2 August, two days before this launch, covering data governance, documentation and transparency for systems that interact with people. Our explainer covers what the Act requires.

Data protection law reaches the same systems from a different direction, and the erasure and accuracy duties our piece on GDPR and model weights describes are considerably easier to satisfy if you logged what the agent touched.

Others are selling into the same anxiety. A Marc Benioff-backed company emerged from stealth a day earlier promising a step-by-step route to deploying agents in an enterprise, and Skan raised $63 million to map how enterprise work actually gets done.

The question to ask about Ethyca Astralis

Real-time governance means intercepting requests, which means sitting in the path of every call an agent makes. Ask what that does to latency, and what happens when the governance layer itself is down. If it fails open, it isn’t a control.

Ask what it can actually block, too. Preventing an agent from reading a restricted table is tractable; noticing that a permitted read was used for an unintended purpose is a much harder claim.

Products like this keep arriving, and our piece on the remedial tooling economy looks at what that pattern says about how the capability was shipped.

Investors have priced the same gap, with $164 million committed in ten weeks to agent infrastructure rather than to models.

A tooling market growing this fast around a capability is usually a sign the capability shipped before its controls did, and buyers are now paying twice for that sequence.