UK DUAA right to complain enforces in -- days · 19 June 2026 EU AI Act high-risk enforces in -- days · 2 August 2026

For Chief Information Security Officer

The AI you do not know is running.

AI exists in scripts, vendor tools, internal automations, and local inference systems. These operate outside governance. Before OBEXGATE enforces anything, it identifies what is actually there.

If you cannot see it before it happens, you cannot prevent it.

Not visibility. Prevention.

Observer Mode shows how your system behaves

Enforce Mode blocks actions before execution

If an action violates policy, it does not run

The visibility problem

Most security tooling assumes it knows what is running.

That assumption is false. Embedded AI, shadow integrations, and unmanaged inference endpoints sit outside the registered surface. A control that does not see them does not govern them.

OBEXGATE identifies what exists before enforcing it. Discovery feeds the same evaluation engine that runs at execution.

Operational difference. Without discovery: unknown systems run without governance. With discovery: hidden systems are surfaced, classified, and brought into the same control surface as registered agents.

Detail on Shadow AI Discovery.

What it produces for your SOC

Tamper-evident evidence, on the channels you already monitor.

Every governed decision produces a structured record. Verdicts, audit lineage, and incident artefacts route to the SIEM you already operate.

Surface What it does
Decision trace Per-action record: what was evaluated, which frameworks applied, why the decision was made
Audit lineage Tamper-evident chain across every governed event. Cryptographically verifiable.
SIEM egress Splunk, Datadog, Elastic, SentinelOne, plus custom HTTP endpoints
Drift detection Continuous, async monitoring of behavioural shift across agent population. Alerts route to your incident pipeline.
Contestation workflow Right to challenge automated decisions. Regulatory requirement under EU AI Act and several jurisdictions.

Adoption pattern

Observer Mode first. Enforcement when verified.

Every paying tier ships with Observer Mode. The same evaluation engine runs, but verdicts surface as alerts rather than blocks. This produces the data your team needs to validate rules before enforcement is enabled, and matches the testing cycle most enterprise security programmes already run.

When verification is complete, enforcement activates without redeploy.

Core invariants

Properties the system enforces, not policies you configure.

→ No execution without evaluation

Every governed action passes through the verification gate. No bypass path exists in the topology.

→ Unknown agents surfaced

Discovery runs continuously. Newly observed agents are classified before they are permitted to operate.

→ Enforcement cannot be bypassed

The commit authority is runtime-owned. There is no path to mutate state outside the gate.

→ Every decision traceable

Decision provenance is produced as a side effect of operation. Not assembled later.

See what is running that you have not registered.

Six questions. Personalised regulatory map, cost basis, statutory exposure. Or 30 minutes with the team to walk through what discovery would find in your environment.