Private AI · SOC Controls

SOC-grade control over your AI agents

Deploy autonomous agents under the same control framework that governs the rest of your regulated infrastructure — data confined to your perimeter, every action authorized and logged, evidence ready for the auditor.

  • Data never leaves your perimeter
  • Policy-enforced action layer
  • Tamper-evident audit trail
  • SOC 2-aligned evidence
0
bytes of your data sent to third-party APIs by default
5
Trust Services Criteria mapped end to end
100%
of agent actions logged with full lineage
<60s
to halt the fleet with a scoped kill switch
// confidentiality

Your data stays inside your perimeter

The fastest way to fail a security review is to discover that a vendor's "AI agent" is quietly POSTing your records to an endpoint you don't control. We design the opposite: the entire stack — models, orchestrator, vector store, prompts, and logs — runs inside your VPC, on your own hardware, or fully air-gapped.

When an external model genuinely earns its place, egress runs through a broker you own. You whitelist the destination, redact or tokenize the fields that go out, and capture an immutable record of exactly what crossed the boundary.

  • Self-hosted models and vector stores
  • Brokered, field-level egress with redaction
  • Air-gapped option for the most sensitive workloads
// the control plane

Controls that satisfy an auditor

Each control maps to a Trust Services Criterion and produces evidence you can hand over without a fire drill.

// how we get you there

From workload to audit-ready

A measured path that puts the controls in before the agents touch anything that matters.

01

Scope

Define the trust boundary, classify the data in play, and map each requirement to a Trust Services Criterion.

02

Harden

Stand up isolated compute, identities, secrets, and the policy-enforcing action layer inside your perimeter.

03

Instrument

Wire tamper-evident logging into your SIEM, SSO into your IdP, and approvals into your on-call workflow.

04

Attest

Generate the access reviews, change records, and audit trail your assessor needs, then operate under continuous monitoring.

Convenient AI vs. controlled AI

The difference shows up the moment your security team asks where the data went.

A typical hosted agentAn Automatic.co controlled deployment
Data locationVendor's multi-tenant cloudYour VPC, hardware, or air gap
EgressOpaque, always-onBrokered, whitelisted, redacted
IdentityShared API keyScoped identities via your IdP
Audit trailPartial vendor logsHash-chained, streamed to your SIEM
Compliance"Trust us"SOC 2-aligned, evidence-ready

Frequently asked questions

Does our data leave our environment at any point?

No. Agents, the orchestrator, prompts, embeddings, and logs all run inside your VPC or on your hardware. Nothing is sent to a third-party API unless you explicitly whitelist one — and even then, only the fields you approve, through a brokered egress you control.

Are these controls aligned to SOC 2?

Yes. We map the deployment to the SOC 2 Trust Services Criteria — Security, Availability, Confidentiality, Processing Integrity, and Privacy — and produce the access reviews, change records, and tamper-evident audit trail your auditor expects as evidence.

How do you stop an agent from doing something it shouldn't?

Every tool call passes through a policy-enforcing action layer. Scopes, rate limits, and risk thresholds are enforced per agent; high-impact actions require human approval; and a kill switch can halt the fleet in seconds while preserving full lineage.

Can your controls plug into our existing SIEM and SSO?

Yes. We emit structured logs to your SIEM, authenticate through your IdP via SSO/SCIM, store secrets in your vault, and raise incidents into your existing on-call workflow. We instrument your stack rather than asking you to adopt ours.

Pass the security review on the first pass.

One working session to map your trust boundary and the controls that get agentic AI into production without a compliance standoff.