Microsoft’s reported insider-threat comparison concerns the authority organisations give agents over sensitive data and tools. It is a vendor’s framing of a security problem, not a finding that AI software has the motives of a human insider.
AI agent permissions and system controls
The National Cyber Security Centre (NCSC) emphasises that established cyber security fundamentals remain relevant when implementing machine learning models. Organisations are encouraged to extend basic principles to account for specific machine learning risks. This includes maintaining robust supply chain security, implementing appropriate access controls, and prioritising user education. These measures form part of the NCSC's broader framework for the security of machine learning systems.
Identifying Potential Security Challenges
TechCrunch reports Microsoft’s argument for separating the supply of AI models from the systems that monitor and control their authority. Microsoft also sells agent-security products, so its proposed solution should be understood in that commercial context. Separately, NIST’s experiments describe how injected instructions can redirect a tool-using agent into harmful tasks. Neither source establishes a measured incident rate for UK businesses.
Survey barriers are a separate question
The linked UK adoption study used telephone interviews conducted between February and May 2025. It provides historical evidence about reported use and implementation barriers. Those survey findings should not be read as current incident statistics or proof that security concerns alone caused businesses to delay adoption.
Sources and further reading
Microsoft says AI agents are like insider threats as 'frontier models can't be trusted' | TechCrunch
https://techcrunch.com/2026/10/19/microsoft-says-ai-agents-are-like-insider-threats-as-frontier-models-cant-be-trusted/
Thinking about the security of AI systems | National Cyber Security Centre
https://www.ncsc.gov.uk/blog-post/thinking-about-security-ai-systems
Technical Blog: Strengthening AI Agent Hijacking Evaluations | NIST
https://www.nist.gov/news-events/news/2025/01/technical-blog-strengthening-ai-agent-hijacking-evaluations
AI Adoption Research - GOV.UK
https://www.gov.uk/government/publications/ai-adoption-research/ai-adoption-research