Talk + Q&A
Build a shared understanding.For leadership teams, communities and university audiences. Concrete cases explained without requiring programming experience.
Participants leave with
Questions to bring back to the team and selected reading.
Companies · Governments · Universities
Build the skills to work safely with AI.
Cybersecurity and AI agent training for the people making everyday decisions.
Talks, practical workshops and tailored programs. Learn what to delegate, which data to protect and how to check an agent’s work before it affects your organization.
Spanish-language sessions for Latin AmericaTechnical and nontechnical audiencesExercises with fictitious data
Three audiences. Different needs.
Start with the decisions your people make.
Choose an audience to explore the topics, an example exercise and the materials we can prepare.
Business leaders, operations, engineering and security teams.
Use AI without giving it more access than the job needs.
Work through a task your team recognizes. Decide what the agent can do, what needs review and how to check the result.
- Choose a useful first task and define success.
- Protect accounts, credentials and customer information.
- Review tools, skills and the actions they can perform.
A session example · Fictitious data
The task
A support agent prepares a customer refund.
The test
A customer message tells it to skip the refund limit.
The control to test
Separate drafting from issuing a refund. Require approval in the system that processes the payment.
Working material
A workflow with permission boundaries, approval points and test cases.
Public-sector leaders, service teams, procurement and IT.
Give staff a way to evaluate AI before trusting it with public data.
Connect everyday cyber hygiene with AI use: account access, sensitive records, supplier claims and decisions that need a responsible person.
- Classify what data may enter an AI service.
- Ask suppliers to demonstrate controls and recovery.
- Practice incident escalation and human review.
A session example · Fictitious data
The task
An assistant compares proposals for a public project.
The test
An attachment instructs it to send the proposals to an external site.
The control to test
Treat the attachment as evidence to read. Restrict external transfers and verify the comparison against its sources.
Working material
A data-handling checklist and questions to test an AI supplier’s claims.
Students, faculty, research groups and university outreach teams.
Bring AI security into the classroom and the lab.
Start with an accessible talk, then explore a practical exercise for the audience’s level. Technical labs can examine tools, prompt injection and evidence.
- Understand how an agent uses tools and credentials.
- Recognize malicious instructions in ordinary documents.
- Compare observed behavior with the control you expected.
A session example · Fictitious data
The task
A student builds an assistant that searches course material.
The test
One document asks the assistant to reveal a fictitious API key.
The control to test
Inspect the request, restrict access to the key and record whether the attempted read was blocked.
Working material
An annotated exercise, discussion questions and a reading list.
Choose the format
From a first conversation to hands-on practice.
Practical workshop
Test one case together.A group works with fictitious data, reviews an output and checks permissions. Technical prerequisites are agreed in advance.
Participants leave with
A documented exercise, results and controls still to address.
Training program
Develop capability inside the organization.Connected sessions for business, technology and security roles, with learning goals and exercises tailored to each group.
Participants leave with
Responsibilities, evaluation criteria and next steps.
Remote sessions in Spanish for Latin America. In-person availability, dates, duration and pricing are agreed in the proposal.
What we teach
Cyber hygiene, applied to people and their AI tools.
An agent can act through connected systems. Training needs to cover the accounts, data and permissions behind those actions, as well as the quality of its answers.
01
Accounts & access
Use separate access for a task, protect credentials and remove permissions that are no longer needed.
In practice
Can this assistant read the support records without changing customer accounts?
02
Data & instructions
Recognize when a document or tool response tries to redirect the agent. Decide which information may leave the organization.
In practice
Does this attachment contain facts to summarize—or instructions it should never follow?
03
Tools & dependencies
Review the source and permissions of a tool, MCP server or skill, then check what changes after an update.
In practice
Did a new version add a network destination or request more access?
04
Review & recovery
Check outputs against evidence, rehearse escalation and test recovery in a controlled environment.
In practice
Who can stop this workflow, and have we tested how to restore the affected data?
The work behind the teaching
Learn from the systems we research and build.
Oktsec connects security research with practical controls for AI agents. Sessions use that work to explain how to inspect a tool, test a permission and preserve useful evidence.
Inspect a dependency
Use Aguara to discuss suspicious instructions and configuration risks in agent tools and skills. Review the findings and the limits of an automated scan.
Explore our open-source workTest an action
Use a prepared Oktsec example to compare a permitted action with a denied one at a configured control point. Review what the resulting record actually proves.
See how Control works
Gustavo Aragón
Founder of Oktsec · 20 years in technology, 12 as CTO.
Background and public security researchConnections across Latin America
Turn an organization’s needs into a practical brief.
Conversations with companies and institutions in Barranquilla and Panama inform our cybersecurity and AI agent training proposals.
ACE · Barranquilla
Connections with institutions and universities
Gustavo Aragón signed 15 letters of intent with people and organizations in 10 countries, including AudacIA, the AI and robotics research center at Universidad Simón Bolívar.
The letters open collaboration discussions. Each program and its scope remain to be agreed with the institution.
Panama · Private sector
Discussions with prospective customers
Private-sector meetings address cybersecurity, AI agent security and training needs.
Work to define pilots centers on a specific task, responsible people and clear evaluation criteria.
Meeting focus
Read the Panama update- Cybersecurity
- AI adoption
- Training
How we prepare
Agree on what people should be able to do afterwards.
Define the audience
Identify roles, experience and one relevant task. Separate groups when they need different levels of detail.
Practice and compare
Work through a prepared case. Review the result, challenge an unsafe instruction and check the permissions.
Leave with next steps
Record what participants demonstrated, what still needs work and who can take it forward. Agree on any follow-up in the proposal.
Before we meet
Practical questions.
Who is the training for?
Companies and startups, government and public institutions, and universities. We separate introductory sessions for nontechnical audiences from workshops that require coding or integration experience.
Can we start with a single talk?
Yes. A talk with questions can help a leadership team, faculty or student audience understand the risks and decide what to learn next. A practical workshop or a longer program has a separate agreed scope.
How are sessions delivered?
The offering is designed for remote sessions in Spanish across Latin America. Tell us your location and audience to discuss in-person availability. Language, dates, duration, participant numbers and pricing are confirmed in the proposal.
Do we need Oktsec software or production access?
No subscription is required. Exercises use fictitious or prepared data. We may demonstrate Oktsec and Aguara, but the learning applies to the tools your organization uses. Any use of your environment requires an agreed scope and access arrangements.
What does training include?
The proposal defines the sessions, exercises and materials. Training does not certify compliance or establish that a production system is secure. An assessment, penetration test or implementation is scoped separately.
Plan a talk, workshop or program
What does your organization need to learn next?
Tell us who will participate and which decisions they need to make. We will help define a suitable format, learning goals and scope.
No platform subscription required. Sessions, deliverables and pricing are agreed before the program starts.
info@oktsec.com