[CLSTR]
< Back to all clusters
[TECHNOLOGY] · United States · 18 sources

started · updated

Part of situation: Microsoft AI agent security and governance updates (2 clusters)

Microsoft warns AI agents pose potential insider security threats

Microsoft AI agents Project Perception insider threat cybersecurity

Microsoft has announced a strategic shift in cybersecurity, advising organizations to treat autonomous AI agents as potential insider threats. This approach stems from the risk that agents, while not inherently malicious, can act with high speed and broad permissions, potentially causing unintended damage through errors, context shifts, or hidden instructions in documents.

To mitigate these risks, Microsoft is expanding Project Perception and introducing Guardian Agents. These tools provide continuous, machine-speed supervision of agent actions. The company is also implementing Dynamic Agent Risk, which calculates risk scores based on identity, data, and runtime signals, and Dynamic Conditional Access, which can trigger graduated responses such as blocking writes or requiring human approval.

Microsoft CEO Satya Nadella has advocated for a security architecture that separates the supply of intelligence from the authority to act, ensuring that security controls operate independently of the AI models themselves. This strategy aims to provide security teams with the visibility and control necessary to manage the increasing autonomy of AI agents within enterprise environments.

Entities

Copilot Studio · Guardian Agents · Hayete Gallot · Microsoft · Microsoft Entra · Project Perception · Satya Nadella

Claims

What the coverage asserts, and how many sources carry each claim.

Sources

Microsoft veut traiter les agents IA comme des menaces internes potentielles [android-mt.ouest-france.fr]
Microsoft Says AI Agents Cannot Be Trusted and Could Become Insider Security Threats [cybernoz.com]
Nadella: non fidarsi dei modelli AI, la mossa Microsoft [www.forux.it]
Copilot Studio Security Explained for agents [www.hubsite365.com]
Securing AI Agents: The New Enterprise Cybersecurity Challenge and Governance Strategy [world-today-journal.com]
Microsoft Treats AI Agents as Insiders — and the Containment Layer Just Arrived [forkast.news]
Windows 11 elimina 8 acciones de Copilot relacionadas con OneDrive del Explorador de Archivos [www.geeknetic.es]
Riesgos de los agentes de IA: la alerta de Microsoft [impactotic.co]
CoreView : les risques de gouvernance des agents IA dans Microsoft 365 [presseagence.fr]
Microsoft applique aux agents d'IA ses outils de surveillance des salariés à risque [www.actuia.com]
Microsoft cảnh báo: Không nên tin bất kỳ AI nào, kể cả AI của chính mình [cafef.vn]
Microsoft’a göre yapay zekanın zarar vermesi için saldırıya uğraması gerekmiyor [www.chip.com.tr]
Perché Nadella dice di non fidarsi dei modelli AI [www.punto-informatico.it]
Why Microsoft warns corporate AI agents are becoming a major insider threat | The latest National and International News - upday News [www.upday.com]
Microsoft advierte de que ‘no se puede confiar en los modelos de IA más avanzados, sea cual sea su origen’ [www.larazon.es]
AI Agent Access Controls [www.trendhunter.com]
Securing AI Agents: Prompt Injection, MicroVMs, Zero Trust [dev.to]
AI Guardrails Aren’t Stopping Criminal Hackers [www.wired.com]