What are multi-agent AI systems?

These are systems where multiple autonomous AI agents interact to accomplish tasks. Such systems can scale to millions of agents, introducing new security risks.

What threats do multi-agent AI systems pose?

Experts fear escalation of cyberattacks, fraud, and other malicious activities that could be automated and scaled using AI agents. For example, prompt injection attacks or self-replicating malware may emerge.

Why is Google DeepMind investing in this area?

The company aims to proactively address potential risks from mass AI agent adoption and stimulate academic research in system safety before it’s too late.

← All news

Artificial intelligence

Google DeepMind Invests $10M in Multi-Agent AI System Safety

June 12, 2026

Photo: MIT Technology Review

Quick answer

Google DeepMind and partners have allocated $10 million to study the safety of multi-agent AI systems, aiming to prevent risks from mass adoption of autonomous agents, such as cyberattacks and fraud.

Google DeepMind has announced a $10 million grant program to study the safety of multi-agent artificial intelligence systems. The initiative includes partners such as Schmidt Sciences, the UK’s ARIA agency, the Cooperative AI Foundation, and Google.org’s charitable arm. The funds will support academic research, where experts believe long-term risks can be analyzed more thoroughly than in corporate labs.

James Fox, a director at Schmidt Sciences, and a DeepMind representative emphasized that a dedicated field for multi-agent AI safety research does not yet exist. However, as the number of autonomous agents interacting with each other grows, the risk of unpredictable scenarios increases. For instance, AI agents could amplify existing threats like cyberattacks or fraud by automating and scaling them.

Experts compare potential risks to challenges arising from human interactions in large institutions, but with AI agents, consequences could be more rapid and far-reaching. DeepMind estimates only a few months remain to prepare for the mass adoption of such systems. The grant program aims to accelerate the development of a scientific foundation to prevent potential threats.

Common questions

What are multi-agent AI systems?: These are systems where multiple autonomous AI agents interact to accomplish tasks. Such systems can scale to millions of agents, introducing new security risks.
What threats do multi-agent AI systems pose?: Experts fear escalation of cyberattacks, fraud, and other malicious activities that could be automated and scaled using AI agents. For example, prompt injection attacks or self-replicating malware may emerge.
Why is Google DeepMind investing in this area?: The company aims to proactively address potential risks from mass AI agent adoption and stimulate academic research in system safety before it’s too late.

Dzen feed: /feed/dzen.xml · RSS: /feed.xml