Google Cloud predicts AI agents will reshape cyber attack and defence in 2027

Detailed summary

Annual security forecast warns of a shift from data theft to “decision manipulation,” the death of the patch queue, machine-speed attacks, and ungoverned “agent sprawl” as the...

GUEST RESEARCH: Annual security forecast warns of a shift from data theft to “decision manipulation,” the death of the patch queue, machine-speed attacks, and ungoverned “agent sprawl” as the next shadow IT

Google Cloud has released its security forecast for 2027, predicting that autonomous AI agents will transform how organisations are attacked and defended, and that security teams must fundamentally change how they build, trust and govern these systems.

In the forecast, written by Google Cloud Chief Information Security Officer Chris Betz, the company says agentic AI is now operational at scale, used to find and fix vulnerabilities, with software shipping in hours or days rather than weeks or months. It sets out how both attackers and defenders will use this shift.

The forecast argues defenders hold a key advantage in their internal context, including identity privilege graphs, network architecture, code lineage and runtime data, which can be used to find dangerous permission combinations attackers cannot see. It says every AI agent should be treated as a first-class identity with verifiable credentials and least-privilege access.

Google Cloud also warns that single, generalist AI agents will fail at complex security work. Instead, it recommends coordinated teams of specialised agents, such as hunter, decoder and responder agents, operating with graduated autonomy and a central coordinator that knows when to escalate to a human. It adds that adversarial red-teaming of agentic systems should be a mandatory step before deployment.

The forecast was developed with contributions from Google Cloud and Google security leaders including Karthik Narain, Ramya Chitrakar, Heather Adkins, Sandra Joyce and Charles Carmakal.

Opening of the article, excerpted from iTWire. To read the rest, open the article on the publisher's site.