AI Agents Pose Growing Security Risks Amid Rapid Adoption

A new report from SailPoint highlights the rapid adoption of AI agents across organizations, alongside significant security gaps. Titled 'AI agents: The new attack surface', the survey reveals that 82% of organizations already use AI agents, but only 44% have policies to secure them. Despite 96% of tech professionals viewing AI agents as a growing risk, 98% of organizations plan to expand their use in the next year.

Key Findings:

• AI agents (or agentic AI) are autonomous systems that perceive, decide, and act to achieve goals, often requiring multiple machine identities to access data and services.
• 72% of organizations believe AI agents pose a greater security risk than other machine identities.
• Top concerns include:
  • Access to privileged data (60%)
  • Unintended actions (58%)
  • Sharing sensitive data (57%)
  • Decisions based on inaccurate information (55%)
  • Accessing/sharing inappropriate content (54%)

Security Incidents Already Occurring

• 23% of respondents reported AI agents being deceived into revealing credentials.
• 80% said their AI agents took unintended actions, including:
  • Accessing unauthorized systems (39%)
  • Sharing sensitive/inappropriate data (33%)
  • Downloading sensitive content (32%)

Expert Insight

Chandra Gnanasambandam, SailPoint’s CTO, warned: "Agentic AI is a powerful force for innovation but also a potential risk. These agents operate with broad access and limited oversight, creating a prime target for attackers." He urged an identity-first approach with real-time permissions and least-privilege access.

Call to Action

The report stresses treating AI agents as distinct identities and implementing comprehensive security solutions with:

• Unified visibility
• Zero standing privilege