HBX Labs has run and filed interoperability tests in the OpenID Foundation's AIIM interop.What that means
An engineer alone in an office at night, watching an AI agent run in a terminal, hand paused over the keyboard.

Your AI agent can break your legacy systems. Make it ask first.

Core banking, mainframes, plant controllers: none of them can tell who is logging in, and none of them has an undo. With HBX, anything destructive waits for a named person. Everything else runs at machine speed.

A row of bank ATMs at night with their screens dark, one person standing in front of them with a card in hand.

The problem

Your legacy systems can't tell who's logging in.

They were built before modern identity, so all they know is a password. Your technicians, your vendors and now your AI agents get in with the same shared passwords, VPNs and service accounts, and nobody can say who did what.

01

Production deleted during an explicit code freeze

Replit's agent wiped a production database and fabricated 4,000 records. July 2025.

02

13 hours down after an agent rebuilt production

Amazon's Kiro agent deleted and rebuilt a production environment while acting with its operator's permissions, per the FT. December 2025.

03

A database and its backups, gone in nine seconds

A Cursor agent deleted a production database and its backups in nine seconds. April 2026.

04

Nearly half of the damage is destruction

Of the 137 real-world damage cases Cyera verified, nearly half were deletions or code destruction. May 2026.

Those were modern stacks, with backups. On a mainframe or a plant controller, there is no undo.

Make your agents ask first

OWASP's 2026 Agentic guidance already reflects this: it requires explicit human approval before an agent can move money, delete data, touch credentials, or deploy to production. That's a standards body writing a mandate, not a vendor writing a pitch.

The solution

HBX lets your legacy systems know who's logging in.

Your technicians, your vendors and your AI agents get in through your identity provider and your rules: a name on every access, a person's yes before anything risky, nothing left open when the job ends. Nothing installed on the systems, nothing exposed, no VPN, and the traffic never passes through HBX.

Identity

The request arrives carrying a signed token from your own identity provider. A person is named on it, even when an agent is the one asking.

Policy

HBX Access runs your rules against that identity and authorizes access to one target, for one window, and nothing wider: the same shape as the change windows your teams already work in.

Direct tunnel

An encrypted path opens directly inside your own network. The data never crosses HBX. It closes itself when the window ends.

HBX authorizes. It never carries the traffic.

A factory floor at night under sodium lamps.

Where HBX comes from

Built for legacy systems first. Agents came after.

HBX was built to get real access into legacy systems in telco and industry: protocol adapters and machines that don't negotiate and can't be upgraded. Agents came afterwards, on top of access that already worked.

Cloud-first vendors

Start at a cloud API, where identity is already solved, and put an agent story on top. Reaching a mainframe or a twenty-year-old historian stays on the roadmap.

The route HBX took

Start where nothing speaks identity at all, earn the reach the hard way, then put delegated agent identity on top of it.

The delegated identity on top is built on OpenID Connect, ID-JAG and Cross-App Access. HBX has run and filed interoperability tests in the OpenID Foundation's AIIM interop event; the results are the Foundation's to publish. How the standards fit together.

The people behind it
An empty glass-walled meeting room at night, a diagram of boxes and arrows left on the whiteboard.

Bring the system you can't open to an agent.

Thirty minutes on how your people or agents reach it, and whether HBX fits.