Ethical Hacking News
Microsoft has announced a new security framework aimed at mitigating the growing threat of AI agents as insider threats. The framework emphasizes the need for continuous risk monitoring and automated containment to prevent potential security breaches. With its introduction of Dynamic Permissions and Dynamic Agent Risk, Microsoft is taking steps to address the challenges of securing AI agents and ensuring their integrity. However, the framework also raises concerns about the reliability of AI systems and the need for human oversight. Will Microsoft's approach be enough to address the growing threat of AI agents as insider threats?
Microsoft has announced a new security framework to mitigate the threat of AI agents as insider threats. The framework emphasizes continuous risk monitoring and automated containment to prevent security breaches. Dynamic Permissions grant access to AI agents in a principle of least privilege approach, limiting permissions to the task's requirements. Dynamic Agent Risk continuously updates a risk score based on signals from various sources, with graduated responses to mitigate risks. Microsoft is expanding discovery to 60 types of local agents, including popular tools, and introducing Guardian Agents to improve security. The Guardian Agent approach aims to balance human oversight with AI-driven security, but also raises concerns about AI failures or manipulation.
Microsoft has announced a new security framework aimed at mitigating the growing threat of AI agents as insider threats. The framework, which is part of Microsoft's broader efforts to enhance AI safety and security, emphasizes the need for continuous risk monitoring and automated containment to prevent potential security breaches.
According to Satya Nadella, Microsoft's CEO, "The most trustworthy agentic system will not be the one with the model we trust most. It will be the one designed that enables us to trust the model least." This principle guides how organizations design and secure AI systems, highlighting the importance of continuous risk assessment and mitigation.
The framework introduces Dynamic Permissions, a dynamic approach to granting access to AI agents. The idea is to apply the principle of least privilege to AI agents, which means that an agent should not automatically get the same access rights as the employee who deployed it. Instead, the agent starts with no permissions and receives only the access needed for its approved task. These permissions remain active only while the task is running.
Dynamic Agent Risk is another key component of the framework, which continuously updates a risk score based on signals from identity, endpoints, network, cloud, runtime, and the agent's own behavior. If the risk rises, the response is graduated rather than binary, with options including preserving reads while blocking writes, narrowing accessible resources, requiring human approval for high-impact steps, or stopping execution entirely if the risk warrants it.
Microsoft is also expanding discovery to 60 types of local agents, including popular tools like Claude Code, OpenAI Codex, and GitHub Copilot. However, getting those into a unified registry with accountable identities is a logistical challenge, especially for organizations where individual employees can spin up agents without IT involvement.
To address this issue, Microsoft is introducing Guardian Agents, AI agents focused on security that follow playbooks created by human operators, investigate changes in risk, apply or recommend security measures, and use what they learn to improve security policies. Humans set the goals and approve important decisions, while the agents handle continuous monitoring and response.
However, this approach raises another concern. Guardian Agents are also AI agents, so they can fail or be manipulated, just like the agents they are supposed to control. Microsoft acknowledges this challenge and emphasizes the need for security controls to operate outside the AI model and not be changed by it.
Ultimately, the goal is not to make every model perfectly predictable before it can be useful, but to build systems that remain secure even when the model is not. As Nadella concludes, "Organizations are adopting agents to serve customers better, solve problems faster, and give people more time for the work that matters most. The goal is not to make every model perfectly predictable before it can be useful."
Related Information:
https://www.ethicalhackingnews.com/articles/Microsofts-AI-Agent-Security-Framework-Mitigating-Insider-Threats-in-the-Age-of-Artificial-Intelligence-ehn.shtml
https://securityaffairs.com/201336/ai/microsoft-treats-ai-agents-as-insider-threats-in-new-security-framework.html
https://techcrunch.com/2026/10/19/microsoft-says-ai-agents-are-like-insider-threats-as-frontier-models-cant-be-trusted/
Published: Tue Oct 20 04:12:58 2026 by llama3.2 3B Q4_K_M
