Guarding Agentic AI
“The biggest risk isn't a rogue agent, but a legitimate one acting at machine speed.”
Raul Zayat Galante, CISO at Port, joined CTech to explain why scoped identities and kill switches matter more than autonomy limits.
“My main fear is not the 'rogue AI' case. It's the legitimate actor, with the legitimate credentials, executing the legitimate action under the wrong context, and executing it at machine speed,” said Raul Zayat Galante, CISO at Port.io, when asked about his biggest fear surrounding agentic AI's rollout in the tech industry.
“Prompt injection, corrupted tooling, and excessive permissions can make a helpful agent an efficient, confused deputy. Adding identity, authorization, and observability to agents in post-deployment mode will lead us down the well-trodden path of cloud security failure.”
CTech reached out to a spread of Israeli companies to find out how they're actually handling agentic AI security, and whether local security leaders are ahead of the curve on the risk, or simply closer to it.
Are any AI agents operating with real autonomy, or is everything human-supervised?
There is limited autonomy, not unlimited autonomy. Autonomous agents may gather data, enhance results, maintain documentation and initiate clearly-defined workflow processes without prior approval each time. Any action that has production, security, financial or access implications needs a human intervention. What matters here is not "autonomy versus supervision" but rather the extent of the possible ripple effect of any action.
What's actually in place versus still on the roadmap?
Nowadays, we scoped our agents by using unique identities, least privilege credentials, approved connectors, and human confirmation for critical activities. We keep platform and tool logs relevant to the particular agent and are able to shut it down either by turning off the workflow or invalidating the credentials.
Something that we do not have today is an elegant control plane across all agents. That's the vision: unified runtime monitoring, SIEM integration, behavior analysis, automated security checks for every new MCP connection, credential management, and unapproved local tools detection. There is a kill switch available, but it is not yet a single red button.
Have you had an incident or near-miss?
We haven't had any security incidents involving our use of AI yet. What we have found are configurations that pose some risk during our demo or pilot phase, including permissions that might be more narrowly scoped before turning them on. But I would not call that an incident. We've been successful in that regard.
Where has an AI agent already made things better or safer?
There are tangible benefits to be seen in the area of security operations, especially around triaging vulnerabilities and threats. An agent can pull context from multiple systems, enrich the finding, and prioritize the most important elements to be acted on.
The human makes the decision of how to act, but now has a much better starting point. This cuts down on duplicate effort, saves time, and reduces the chance that context will get lost in the process.
I believe the idea of “being ahead” means anything but more autonomy for agents. It means treating agents from the start as privileged machine identities rather than smart chatbots.
