Microsoft is building a tool, "Guardian Agents," to monitor
artificial intelligence (AI) systems running inside a business's network as part of its "Project Perception" security strategy.
Guardian Agents continuously assess risk across an
entire ecosystem, according to the news, which was announced Monday in a blog post.By integrating the enterprise systems with Microsoft 365, the agents can share signals and context.
This provides IT and security teams with one view of agent activity and risk from cyber attacks, alongside the controls required to manage them.
advertisement
advertisement
“The most trustworthy
agentic system will not be the one with the model we trust most," Satya Nadella, Microsoft CEO, wrote in a post accompanying the announcement. "It will be the one designed that enables us to trust the
model least.”
An agent can fail or be compromised as it becomes an insider with the company, according to the post. This means it is important to minimize and tailor its authority
before it acts, and continuously adjust that authority as risk changes.
Microsoft suggests that companies know their agents when they give them accountability, because it is not
possible to secure what humans cannot see or understand.
Most company executives have focused more on the risks of AI and security outsiders using models to attack.
This specific strategy designed for cybersecurity and enterprise systems does not apply to advertising, but its framework does.
The fundamental principles of discovery, unified registry, first-class identity, and
explicit attribution required across enterprises relates to the future of advertising technology.
As autonomous AI agents have begun to browse the web, make purchase decisions, and complete
transactions on behalf of humans, the advertising industry must completely overhaul the ways it targets, attributes, and measures ads.
Bots inflating ad clicks or video views costs the
industry billions of dollars annually.
By extending identity protocols, advertising networks can enforce strict policy targets and can instantly block anonymous, non-verified "shadow"
agents from clicking on pay-per-click (PPC) ads to ensure that ad budgets are spent on valid traffic.
Microsoft said it will expand discovery of local agents in
Guardian Agents to 35 additional types, bringing the total to 60
supported local agents, including Claude Code, OpenAI Codex, and GitHub Copilot.
These agents can discover software powered by large language models (LLMs) that act autonomously on behalf
of a users. Companies can bring these agents into a registry to understand ownership, activity, and risk.