Warsaw 24:09:2026 Steve Walker
BRAD CENTURION
When the Firewall Is No Longer Enough
Artificial intelligence is evolving for good and for bad.
Today, AI agents are at an awkward age. They are capable, enthusiastic and increasingly autonomous, but they can also be unpredictable. At times they resemble boisterous children ringing the doorbell and running away.
It would be dangerous to assume that this stage will last.
Agents are learning to research, navigate systems, use software, communicate with other agents and undertake sequences of actions with progressively less human intervention. Soon the important question will not be what one AI agent can do. It will be what hundreds or thousands of coordinated agents can achieve together.
That changes cybersecurity fundamentally.
For organisations possessing valuable intellectual property, one response to AI is already emerging: bring intelligence inside the organisation.
Run your own Large Language Model. Ring-fence proprietary information. Build a private data estate. Prevent commercially sensitive knowledge from routinely travelling to external AI platforms.
That is sensible.
But evolution does not occur on only one side of the wall.
While we develop AI to protect and exploit our intellectual property, counterparties can develop agents whose purpose is to discover more data, more relationships and ultimately more intellectual property.
The contest therefore evolves.
And the answer can no longer simply be a firewall.