/
← Voltar ao feed
low

Tenable guidance on shadow AI: MCP servers, agents and local models widen enterprise exposure

1 fonte(s) · 1 atualização(ões) · publicado 22/10/2026 10:00 · coletado há 3 horas

A Tenable blog post for Cybersecurity Awareness Month argues that shadow AI now covers far more than personal chatbots. It includes unvetted MCP servers, AI browser extensions, locally hosted models, coding assistants and autonomous agents holding broad credentials. Citing IBM's 2026 Cost of a Data Breach Report, it says 43% of organizations had security incidents involving shadow AI assets (up from 20%), at an average breach cost of $5.39 million. Tenable's own telemetry found 457 million AI-related security issues across 7,000 customer environments in 30 days, about 62,000 per customer. The post is vendor guidance, not a report of a specific incident, and promotes Tenable's exposure management approach.

Detalhes técnicos

  • IBM reports that almost half of shadow AI incidents caused data loss or compromise, 42% caused operational disruption and 21% triggered regulatory fines.
  • The Cloud Security Alliance describes an uninventoried "invisible layer of AI infrastructure" with no policy enforcement, access control or audit.
  • Cited surveys: 34% of employees knowingly use unapproved AI (Thomson Reuters) and 31% of UK employees do so without permission (Deloitte).
  • Example risks: MCP servers bridging assistants to internal databases, extensions with read access to CRM data, administrative credentials handed to consumer AI agents, and proprietary code or financial data pasted into public AI tools.

Remediação

  • Add shadow AI to security awareness training and talk to employees and business unit leaders about why they use unsanctioned tools, as the UK NCSC recommends.
  • Avoid "block by default" policies. Publish an AI acceptable use policy with approved tools, data handling rules, accountability and vetting criteria (data ownership, encryption, training use, data residency, compliance).
  • Keep a continuously updated inventory of AI assets across endpoints, cloud, pipelines, OT and IoT. Detect misconfigurations and excessive permissions, enforce least privilege, and monitor and block sensitive data flows to AI tools.

Linha do tempo

  1. tenable.com ↗ · pub. 22/10/2026 10:00 · coletado 22/10 10:00 primeiro registro

    Tenable guidance on shadow AI: MCP servers, agents and local models widen enterprise exposure

    ler a análise desta fonte

    A Tenable blog post for Cybersecurity Awareness Month argues that shadow AI now covers far more than personal chatbots. It includes unvetted MCP servers, AI browser extensions, locally hosted models, coding assistants and autonomous agents holding broad credentials. Citing IBM's 2026 Cost of a Data Breach Report, it says 43% of organizations had security incidents involving shadow AI assets (up from 20%), at an average breach cost of $5.39 million. Tenable's own telemetry found 457 million AI-related security issues across 7,000 customer environments in 30 days, about 62,000 per customer. The post is vendor guidance, not a report of a specific incident, and promotes Tenable's exposure management approach.

    Technical Details

    • IBM reports that almost half of shadow AI incidents caused data loss or compromise, 42% caused operational disruption and 21% triggered regulatory fines.
    • The Cloud Security Alliance describes an uninventoried "invisible layer of AI infrastructure" with no policy enforcement, access control or audit.
    • Cited surveys: 34% of employees knowingly use unapproved AI (Thomson Reuters) and 31% of UK employees do so without permission (Deloitte).
    • Example risks: MCP servers bridging assistants to internal databases, extensions with read access to CRM data, administrative credentials handed to consumer AI agents, and proprietary code or financial data pasted into public AI tools.

    Remediation

    • Add shadow AI to security awareness training and talk to employees and business unit leaders about why they use unsanctioned tools, as the UK NCSC recommends.
    • Avoid "block by default" policies. Publish an AI acceptable use policy with approved tools, data handling rules, accountability and vetting criteria (data ownership, encryption, training use, data residency, compliance).
    • Keep a continuously updated inventory of AI assets across endpoints, cloud, pipelines, OT and IoT. Detect misconfigurations and excessive permissions, enforce least privilege, and monitor and block sensitive data flows to AI tools.