Nobody knows which agents are running.
Arora asks it himself: how do you find all the agents that are running? Without an inventory there is no control.
THE LAB · SECURITY HARNESS
Put AI agents to work.
Without losing control.
Agents read email, log in and trigger actions. A security harness defines what they may do, who approves, and how you can trace what happened. We build it with you, step by step.
30 minutes. Personal. No obligation.WHY NOW
“We haven’t built the AI security harnesses that are needed.”
THE SITUATION
Deployed fast.
Rarely secured.
Four points from the conversation
that matter for small teams too.
Agents use your passwords.
Many personal agents log in with your credentials. To a bank, that looks like you. Who is liable when something goes wrong?
“The AI did it” is not an answer.
Especially in finance and healthcare, decisions are expected to be explainable, auditable and verifiable.
Gaps are found faster.
AI helps attackers find vulnerabilities too. Outdated systems and postponed updates become a real risk.
HOW THE HARNESS WORKS
Five layers around every agent.
Not a tool, but a framework of rules, permissions and routines. We implement it with the tools you already use.
- 01
Inventory
Which agents, automations and AI tools are running, with which data and accounts? We make it visible.
- 02
Identity & permissions
Own credentials instead of shared passwords. Each agent gets only the rights its task requires.
- 03
Approvals
Payments, orders, outbound email: the agent prepares critical actions, a human confirms.
- 04
Logging
What did the agent see, decide and do? Traceable for you, your auditors and your customers.
- 05
Detect & respond
Spot unusual behaviour, stop agents, revoke access. With an incident plan you have practised beforehand.
WHO IS IT FOR?
Wherever agents
are allowed to act.
From the first copilot in your inbox
to your own agent product.
AI in your inbox and calendar.
Assistants read email, answer requests and book meetings. We clarify what they can access and what never leaves without approval.
Traceability that survives an audit.
If you handle sensitive client data, you need logs and clear responsibilities before an agent goes live.
Build security in from day one.
Building a product with agents? In the Lab we design permissions, approvals and logging with you before your first customers arrive.
Adopt AI, keep trust.
Clear rules and documented processes are the basis for using AI responsibly.
HOW TO START
Three ways to a harness.
Using agents safely.
A hands-on workshop at the Hub: how agents work, where risks arise and which rules you can apply right away.
Workshops & eventsAll your agents at a glance.
Together we map which agents and access rights exist and prioritise next steps along the five layers.
Request a checkFrom plan to operations.
Set up permissions, approvals, logging and an incident plan, including training and handover to your team.
Book a callWHY DIGIHUB
We live AI.
That’s why we secure it.
From practice.
We use AI and agents in our own daily work. We know where it breaks because we experience it ourselves.
People first.
A harness only works if the team understands it. Training and guidance are always part of it.
Lab, Hub, Fund, Learning.
Experiment in the Lab, connect at the Hub, learn in workshops. And if it turns into a venture: funding and company formation.
On site in Schaan.
A cooperative for digital transformation in Liechtenstein. Reachable in person, In der Specki 3.
GOOD TO KNOW
Questions about
the security harness.
What is a security harness?
A framework of rules, permissions and routines around AI agents: what they may do, who approves, what gets logged and what happens in an incident. The term comes from the current AI security debate; our approach is tailored to SMEs, start-ups and the public sector.
Do we need this if we only use ChatGPT?
As soon as a tool accesses email, files or accounts, or triggers actions, it’s worth a look. In the intro call we work out whether a workshop is enough or a check makes sense.
Does this replace our IT security?
No. The harness adds the questions AI agents raise on top of firewalls, updates and backups. We’re happy to work with your existing IT.
Do you work with specific vendors?
We start from your processes and tools, not from a product. Recommendations depend on your situation.
What does it cost?
The intro call is free. For workshops, checks and implementation you get a quote that fits the scope and team size.
YOUR NEXT STEP
Which agents are
running at your place?
In 30 minutes we get an overview together and show where a harness helps most.
Background: All-In Podcast, episode “AI Cyber Panic, Ken Griffin’s Bubble Warning, The Fall of France?”, 23 Oct 2026, conversation with Nikesh Arora (Palo Alto Networks). Episode on YouTube. digihub.li is not affiliated with the podcast or Palo Alto Networks.