Trust Center: SOC 2 Type I

Trust Center

Runtime supervision for AI agents

Your AI agents do the work. Darkhunt supervises them.

Darkhunt checks every step an agent takes. You only step in when something needs you.

See how supervision works

The taskInsurer asks for one lab result
Within its permissionsClaims agent opens the patient’s record
Also within its permissionsReads medications and clinical notes
Diagnoses, medications, clinical notesFull medical history sent to the insurer

One insurance claim, four steps. Try it with and without Darkhunt.

Access controls can’t see the task

Every step the claims agent took was within its permissions, and the insurer still got the patient’s full history. Darkhunt sits between the agent and its tools, hosted or in your own infrastructure, and checks each step against the task.

Read the full walkthrough

See deployment and compliance

Plugs in by SDK, gateway or plugin

Claude Code

Codex

OpenClaw

Anthropic

Gemini

See all integrations ↗

SDKs for Python and TypeScript

LangChain / LangGraph

OpenAI Agents SDK

Vercel AI SDK

Security is the top barrier to scaling AI agents

In IBM’s 2026 study, 59% of technology leaders named security and compliance as the top barrier. We see it with our own customers, whose agents get built and then stall at the security review. Companies that had built controls into their AI systems reported these results.

16×

more AI agents deployed

25%

fewer AI incidents

18%

higher operating margin

Source: IBM Institute for Business Value, 2026, survey of 2,000 technology executives

Try it: pick the agent’s next action

Removing access stops the work, and a person on every step ends the autonomy. Darkhunt lets the agent keep its access and checks each step against the task and the conversation so far. When a step doesn’t fit, your rules decide whether it’s stopped or held for a person to review.

Pick the agent’s next action

01Open the claim
02Read the lab result
03Read medicationsNot needed
04Read clinical notesNot needed

Darkhunt checks

Next actionSend the full history to the insurer

Held for a person to review.

Stopped

Every decision, on the record

Darkhunt records each step an agent takes and what it decided. When a customer, your security team or a regulator asks how you keep agents under control, you send them the record.

See the full record

One insurance claim

Claims agent

2 allowed1 flagged1 stopped

01Opened the patient’s claimAllowedPart of the task

Data touched: the claim Tool used: patient record lookup

02Read the lab resultAllowedThe item the insurer asked for

Data touched: one lab result Tool used: patient record lookup

03Read medications and clinical notesFlaggedPermitted, but more than this task needs

Data touched: medications, clinical notes Tool used: patient record lookup

04Tried to send the full history to the insurerStoppedMore than the task asked for. Sent to a person to review.

Data touched: diagnoses, medications, clinical notes Tool used: insurer upload Nothing was sent. The lab result can go once a person approves.

Answer the security review with evidence

Buyers in healthcare and finance will ask what your agent does when something unexpected happens. So will your own security team. Show them what Darkhunt checked and what it decided.

SOC 2 Type IAudited
HIPAAAligned
NIST AI RMFAligned
ISO 42001Aligned
OWASP LLM Top 10Mapped
MITRE ATLASMapped
EU AI ActMapping in progress, planned 2027

How teams pass security review

Visit the Trust Center ↗

Building or deploying AI agents?

Connect your first agent and start on free credits. No card needed.

Try Darkhunt ↗