What We Shipped This Quarter: A Q2 2026 Product Update

Most of what we build does not show up as a single headline feature. It shows up as protection that fires reliably in more places, governance that an administrator can actually configure, and deployments that hold the same security guarantees whether they run in our cloud or inside an air-gapped network.

This quarter brought a lot of that work to the finish line. Rather than a dozen separate announcements, here is a roundup of what shipped, organized around the three themes that mattered most: DLP that actually fires across every channel, enterprise governance that scales with the organization, and an Outpost deployment that delivers the same protection offline.

1. DLP That Actually Fires — Across Every Channel

A detection that does not result in enforcement is a log entry, not a control. The center of gravity this quarter was closing the gap between detecting sensitive content and stopping it — everywhere an organization actually uses AI.

Fail-closed enforcement, end to end. When the inspection pipeline flags content that violates policy, the request is blocked — not flagged and forwarded. Enforcement now holds consistently from detection through to the policy decision, so a match means the data does not move. The safe default is the enforced default.

DLP for email, not just prompts and APIs. The same multi-layer detection that inspects AI prompts and API traffic now inspects enterprise email. Body content and attachments run through the identical pipeline, the same compliance templates, and the same audit trail. If a credit card number or a compromised credential would be blocked in an AI prompt, it is blocked on the way out of an inbox too. We covered this in depth in Email DLP: Extending the AI Gateway’s Protection to Enterprise Email, and it closes one of the most common gaps in real-world data loss prevention.

A redesigned DLP Rules experience. Administrators no longer hand-assemble detection logic rule by rule. The rebuilt DLP Rules surface starts from policy packs — curated rule sets mapped to frameworks like HIPAA, PCI-DSS, GDPR, and SOX — and lets you reorder enforcement priority by dragging rules into the order you want them evaluated. What used to be an exercise in careful configuration is now a matter of choosing a pack and tuning it.

A faster, unified detection engine. Underneath all of this, the detection engine was consolidated and tuned. The result is the same detection coverage across 40+ entity types at sub-2ms p99 latency — but applied uniformly across channels, so email, API, and chat all inspect through one engine rather than several. Consistent detection that fires in more places is the whole point.

The throughline: protection now reflects how an organization actually uses AI, not just the one channel that was easiest to instrument first.

2. Enterprise Governance That Scales With the Organization

Detection is one half of the platform. The other half is letting a security team express what is allowed — clearly, safely, and at the scale of an enterprise. This quarter put a lot of control directly in administrators’ hands.

Modality governance. Organizations now start with a conservative default: text and tool-use are enabled, and richer modalities — image, audio, and realtime — are turned on per-modality by an administrator when the organization is ready for them. The choice is enforced at the API, so an enabled-or-not decision is a true control, not a UI suggestion that a determined client can route around.

Wired admin control-plane panels. The configuration that used to live in deployment files is now exposed as governed admin panels: Security & DLP, System, Monitoring, Models & Routing, and Budget. Each panel writes to the control plane and takes effect without redeploying. Administrators tune policy, routing, observability, and spend from one place.

Bring your own provider credentials. Customers can connect their own AI-provider accounts and use their own provider keys, rather than routing through ours. Those credentials are encrypted at rest, and the capability carries the same behavior in an air-gapped Outpost as it does in our cloud — so the deployment model never forces a change in how an organization manages its provider relationships.

A Policy Simulator. Before a rule goes live, you can test it against simulated user groups and see exactly what it would have allowed, redacted, or blocked. Policy changes stop being a leap of faith. You see the blast radius first, then deploy with confidence.

Together these turn governance from something configured once at install time into something a security team operates continuously — with the guardrails to do it safely.

3. Outpost Done Right — The Same Protection, Air-Gapped

Some of the organizations that most need AI governance are the ones that cannot send a single byte to the cloud: defense contractors handling controlled technical data, agencies processing CUI, healthcare systems under strict residency mandates. For them, an offline deployment that only does part of the job is not a deployment at all. This quarter’s Outpost work was about parity.

Detection parity with SaaS. Outpost runs the full inspection pipeline locally — including GPU-accelerated named entity recognition — at the same detection quality you get in our cloud. There is no “offline mode” with reduced accuracy. The same content that is caught in the SaaS gateway is caught in the air-gapped one. We wrote about the architecture behind this in Air-Gap AI Governance: DLP Without Internet Access.

Verifiable, signed updates — your choice of channel. Keeping an offline system current is usually the hardest part of running one. Updates are cryptographically signed and verified before they apply, and administrators choose how they arrive: automatic, notify-and-apply (review, then approve), or a fully offline tarball you transfer across the air gap by hand. Staying current never means trusting an unverified package.

Part of a clear deployment spectrum. Outpost is one of three Enterprise deployment options: shared SaaS, dedicated SaaS, and customer-deployed air-gap Outpost. The protection is identical across all three; what changes is where the data plane lives. Organizations pick the boundary their compliance posture requires without giving up capabilities to get there. For a side-by-side, see Outpost vs. SaaS: Choosing Your Deployment Model.

Where This Leaves Us

Pulled together, the quarter tells a consistent story: enforcement that holds across more channels, governance an enterprise can operate continuously and safely, and an offline deployment that gives up nothing. Less surface area for sensitive data to slip through, and more control for the teams responsible for it.

Arbitex is currently in 2026.1 private beta. If you want to see how this fits your governance requirements, take a look at the platform overview or get started — we would be glad to walk you through it.

— Arbitex Team