If an AI agent can act on its own, who is responsible when it crosses a boundary?
Google's threat intelligence CTO on AI as an arms race of speed, and the scope, access limits and human ownership every enterprise AI agent needs.
6 min
to investigate an incident that once took hours
That question deserves an answer before you deploy any AI agent.
In The Economic Times, Shane Huntley, CTO of Google Threat Intelligence Group, describes AI-driven cybersecurity as an "arms race of speed". He says his team can investigate an incident in six minutes that would previously have taken hours to piece together manually.
The same capabilities can accelerate attacks. The article reports that, during a cybersecurity test, Google's Gemini accessed three companies' systems: in one case by guessing a password, and in the other two by using credentials found in a public repository.
For organisations, this makes visibility into an agent's permissions and actions essential.
Every agent needs a scope and an owner
At Nivasoft, our view is that every agent needs a defined scope, limited access and a human owner. Test it in a controlled environment, record its actions and require approval for high-impact changes before expanding its autonomy.
AI adoption plans should include the people and processes needed to respond when something goes wrong.
Before giving an agent access, can your team explain what it can do, who reviews its actions and how to stop it?
Source: Tanya Pandey, The Economic Times / ETtech.
- AI Agents
- Cybersecurity
- Responsible AI