Microsoft warns prompt injection can hijack enterprise AI agents
Original title: Microsoft Warns AI Agents Can Be Hijacked Through Prompt Injection Attacks
Project Perception will discover agents, isolate runtimes and apply task-scoped permissions outside the model.
Microsoft warned that enterprise AI agents can take unsafe actions after prompt injection, compromised inputs or task-context shifts, even without malicious intent. An expense-approval agent could be steered by content in an external document while its configuration remains unchanged. The company is extending Project Perception with discovery, separate identities, execution containers and dynamic permissions enforced outside the model.
- Discovery will add 35 local agent types, bringing the total to 60.
- Named types include Claude Code, OpenAI Codex and GitHub Copilot.
- Dynamic Permissions limit authority to a specific task and duration.
- High-risk controls can isolate an agent runtime or stop it.
- Vendors
- Microsoft
From the original article
2 min read at gbhackers.com
Microsoft has issued a warning about the risks posed by AI agents that access enterprise systems, especially when attackers manipulate them through methods like prompt injection, compromised inputs, or shifts in task context. To address this issue, the company is enhancing Project Perception with controls to identify agents, limit their authority, and manage unsafe behavior […]
By Divya · gbhackers.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.