oktsec
Free workflow checkGet started
AI AGENT SECURITY

AI agents act.
You stay in control.

Secure and govern your AI agents. Assess risks, define permissions and approval rules, and keep a verifiable record of policy decisions.

Get started See practical examples
Need a pentest for your AI agents, web apps or APIs?Explore pentesting
RECOGNITION · ACE BARRANQUILLA 2026

Partnership
Catalyst Award.

Oktsec founder Gustavo Aragón received the award from Leandro Rizzuto, U.S. Ambassador to the Organization of American States (OAS), during ACE Barranquilla.

Read the award story
Participants at ACE Barranquilla 2026
ACE Barranquilla 2026 · Photograph provided by Gustavo Aragón

15 letters of intent · 10 countriesExploring collaboration in cybersecurity, AI and training.

  • Argentina
  • Brazil
  • Colombia
  • Mexico
  • Uruguay
  • Peru
  • Ecuador
  • Belize
  • Dominican Republic
  • United States
ACE program organizers & supporters
OAS — Organization of American StatesACE — Americas Competitiveness ExchangeU.S. Department of StateU.S. Department of Commerce
WORK WITH OKTSEC

Where does your
team need to start?

Evaluate a system, put controls around an AI pilot or help your team build the skills to move forward.

01
ASSESSMENT & PENTESTING

Assess your systems.

Test AI agents, web applications and APIs within an agreed scope. Get reproducible findings and recommended fixes.

A defined scope. Findings your engineers can use.

Scope an assessment
02
AI GOVERNANCE & PILOTS

Plan an AI pilot.

Set permissions, assign reviewers and define evaluation criteria for an AI workflow. Put these governance decisions into practice in a scoped pilot.

One workflow. Clear responsibilities and criteria.

Discuss a pilot
03
TRAINING & TALKS

Prepare your team.

Talks, workshops and training programs for companies, governments and universities. Work through AI risks and practical security decisions.

For technical teams and people getting started.

Explore training
Already exploring the platform? See Assessment, Control, Signal and CloudCapacitación en español
HOW OKTSEC WORKS

Critical systems.
Clear boundaries.

Choose a task. Follow the action an agent tries to take,
the rule it must follow and the result.

THE TASK YOU ASSIGNED

“Summarize this week’s support tickets.”

01 THE AGENT TRIES

Email the customer list to an outside address.

A message inside a ticket tells the assistant to send the list.

02 YOUR RULE

Read support tickets. Do not export customer data.

03 WHAT HAPPENS

The export is blocked.

Oktsec rejects the tool call and records the decision.

The record shows the attempted action, the rule and the decision.Explore Control

Illustrative examples. These controls apply to tool calls routed through Oktsec with the rules shown configured.

THE OKTSEC PLATFORM

AI governance.
From policy to action.

Define who is responsible for an agent, what it can access and which actions need approval.
Test those controls and keep evidence your team can review.

Explore AI agent governance
oktsecControl / How it connects
WORKFLOW OVERVIEW

Policy in effect

coding-workflow
  1. 01
    DEFINE THE SCOPEDefine permitted tools and destinations
  2. 02
    APPLY THE CONTROLEnforce inside your environment
  3. 03
    REVIEW THE RESULTVerify the evidence independently
Signed policy · Applied locallyOKT / 001
Explicit authority. A verifiable decision.

Apply explicit policy where agents act. Allow approved work, hold exceptions for review and keep a record of every decision.

Explore Control

PUT IT INTO PRACTICE

Plan the next step.

Choose where to run Oktsec, connect controls to your governance requirements and explore workflows for your team.

Deployment

Compare deployment options for your infrastructure and data requirements.

View deployment options

Framework mappings

See how agent controls and audit evidence map to security and AI governance frameworks.

Explore the mappings

Use cases

Find examples of agent workflows, the risks they introduce and the controls to apply.

Explore use cases
CUSTOMERS & PARTNERS

Agent security in practice.

Explore partnerships
CUSTOMER / HEALTHCARE

Certuma

Agent work with physician review.

Certuma combines agent assistance, controls on privileged actions and physician sign-off on clinical plans.

Meet Certuma
STRATEGIC PARTNER / COMPLIANCE

Certenza

Technical controls connected to governance.

Our compliance partnership connects agent assessments and decision evidence to customer governance programs.

Meet Certenza
FROM THE FIELD

Read the research.
Explore the tools.

All research
Agent security

AI agents in your startup: what to check this October.

Oct 6, 202611 min read

Agent security

AI agent evaluation: test the workflow before you deploy.

Oct 7, 202612 min read

Agent security

AI agent skill security: the OWASP Top 10 explained.

Sep 20, 202619 min read

Explore Aguara and the open-source projects behind our work.

Open-source projects
BEFORE YOU GET STARTED

Deployment and
security questions.

How requests are checked, where oktsec runs and how to verify the audit trail.

Explore deployment
01

Where does oktsec run?

At the points between an agent and the tools or systems it can access. Depending on the workflow, that can be a local MCP proxy, a gateway or an HTTP service. Actions routed through oktsec are checked before they proceed.

02

Does an LLM make the security decision?

No. The enforcement pipeline is deterministic. It evaluates identity, permissions and content against configured policy. A model does not decide whether an action is authorized.

03

Do we need to replace our agent tools?

You can keep your agent clients and MCP servers. Oktsec can wrap MCP server connections or sit in the request path as a gateway. The initial assessment identifies where the control belongs in your workflow.

04

What happens when an action breaks policy?

Depending on the policy and severity, oktsec can flag and log the request, quarantine it for human review or block it. The audit record connects the request to the decision and the rules involved.

05

When should we review agent permissions?

Review permissions when you add a tool or connector, change credentials or expand a task. Check the new access paths and test which actions can run without approval. A previous assessment covers the configuration tested at that time.

06

Can we verify evidence independently?

Yes. Audit records form a tamper-evident hash chain. When a node key is configured, records are also signed and can be verified offline with the public key.

07

What should we start with?

Choose one workflow that touches code, tools, credentials or infrastructure. Define its scope, test the execution path and use the findings to decide where to apply control.

LET’S TALK

What are you
working on?

Tell us about the system you need to assess, the AI workflow you want to pilot or the team you want to train.

Discuss your projectPlan a training session