New·Argos now detects usage deviation across 100+ model endpoints See how →
Home / Security
Security & governance

Our own security posture.

Payers ask about a BAA before the second meeting. Banks ask about SR 11-7. European buyers ask about residency. These are the answers — and the posture behind every agent we ship.

SOC 2 Type II

Independently audited access management, controls and incident response.

Report on request

HIPAA & BAA

PHI stays inside your compliance boundary. Infosec reviews handling before any pilot starts.

BAA capable

Model risk

Model cards, validation evidence, drift monitoring and HITL design your MRM function can review.

SR 11-7 · NIST AI RMF

Data residency & model choice

Your cloud, your region, your models. Vendor neutral, in-VPC deployments.

Multi-cloud

Full audit trail

Every prompt, call, output and approval logged, attributable and reversible.

Regulator-ready

Adverse decisions stay human

No denial, decline, accusation or termination is ever issued by an agent.

Non-negotiable

Citations or it doesn't ship

Every relied-on output carries its source. Unsourced values can't publish or act.

Provenance

Fairness monitoring

Metrics tracked across groups on any agent touching people, thresholds set with compliance.

Continuous

One-action pause

Your team can stop any agent immediately, with no dependency on us.

Kill switch
How autonomy is earned

How an agent earns the right to act.

Stage 0 · before build

Regulatory exposure and control design

The Regulatory Exposure Mapper lists the regulations and controls each use case needs before approve mode. Compliance signs the risk tier.

Stage 1 · shadow

Outputs compared, decisions unchanged

The agent produces outputs; humans decide as before. We measure agreement, coverage and failure modes over a meaningful sample.

Stage 2 · approve

Recommendations, human signature

A named person approves, edits or rejects each output. Acceptance rate and override reasons are tracked and feed the eval suite.

Stage 3 · bounded autonomy

Acts within thresholds the risk owner sets

Value limits, confidence floors, category scopes, volume caps. Everything outside escalates; severe regression rolls back automatically.

Always

Kill switch and quarterly review

Your team can pause any agent in one action. Thresholds reviewed with compliance every quarter.

Ask us for any of these

What we will send your CISO.

Security questionnaire

SIG or your own template. Turnaround in five business days.

BAA template

Our standard, or redlines on yours.

Architecture & data flow diagrams

Per engagement, showing every place data rests and every model it touches.

Model cards & eval reports

For every agent in production, current within the month.

Penetration test summary

Most recent third party test, under NDA.

A call with our CISO

Before a pilot, not after.

Let's talk

Tell us the number you need to move.

A 45-minute working session with an operator who has run the kind of work you are describing. You will get an honest read on where your programme stands and what it would take to move it.