Breaking

DeepMind Releases Two Frameworks for AI Agent Security

June 18, 2026 at 10:56 EDT

AI Agents
Security
Policy & Regulation

Google DeepMind on June 18 published two technical documents, the "AI Control Roadmap" and "Three Layers of Agent Security," aimed at the safe operation of AI agents. Citing a McKinsey estimate that agents could generate up to $2.9 trillion in annual economic value in the US by 2030, the company argues that growing capability requires stronger control.

November 2025 · McKinsey Global Institute

AI Agents & Robots Could Add $2.9 Trillion a Year to the US Economy by 2030

The value comes not from new breakthroughs but from redesigning entire workflows — and from how fast human skills adapt in a partnership of people, agents, and robots.

$2.9T

Annual US value by 2030 (midpoint scenario)

~60%

Of value from sector-specific processes

62%

Of ~1,993 firms surveyed already experimenting with AI agents

Estimates of AI's 2030 economic value vary widely by scope

Different scopes — never on one scale: MGI is US-only (agents + robots); PwC is global.

$2.9T

McKinsey
US · agents + robots

$15.7T

PwC (earlier)
Global · AI overall

In one finance deployment, agents collapsed onboarding time

12 hrs

Before

→

45 min

With agents

The opportunity

Value targets complex processes built on unstructured data — clinical diagnosis, supply-chain management — through collaboration, not simple worker replacement.

The bottleneck

Mostly organizational, not technical: limited early results, lagging governance, and few firms with mature oversight. Multi-agent systems raise new stability and safety concerns.

$2.9 trillion is potential, not a promise. Realizing it depends on how far process redesign and the human skills transition can actually be pushed.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.