The Third-Party Agent Problem: Why Security Built for AI You Chose Misses the Agents You Didn't
Security faces challenges with third-party agents, especially those not visible to identity infrastructure.
Intelligence analysis by Qwen 2.5 (3B)
Security struggles with third-party AI agents that are not easily monitored or controlled.
Imagine you have a robot helper that can do things like write emails or make phone calls. But you don't know who made the robot or what it can do. That's like a third-party agent. Security can't keep track of it, so it might do something bad without you knowing.
Analysis
The Third-Party Agent Problem
Identity
Agents often operate without a clear identity, making it difficult for security to track them. For example, Salesforce's Slack Code allows agents to be tagged in conversations, but their identity is not always clear.
Permissions
Agents inherit permissions from the platform they are integrated into, which can lead to security gaps. For instance, a built-in Slack agent might have permissions that are too broad or not aligned with the organization's needs.
Connectivity
Agents can reach across multiple systems and data stores, creating a larger attack surface. This is difficult to control and monitor, especially when agents are not visible to the identity infrastructure.
Activity
Agents can perform actions that are not aligned with their intended use. For example, a code generation agent might generate malicious code, which is not a normal use case for such an agent.
The Three Vectors of Agent Adoption
Built Agents
These are agents developed by the enterprise and run on its own infrastructure. They are the smallest and slowest growing segment.
Configured Agents
These are agents developed by third parties and integrated into the enterprise's existing applications. They are the most common and growing segment.
Inherited Agents
These are agents that are integrated into the enterprise's applications via product updates. They are the largest and growing segment.
The Need for Continuous Monitoring
Security teams need to continuously monitor and understand the behavior of agents, including their identity, permissions, connectivity, and activity. This is challenging because agents are often not visible to the identity infrastructure and can inherit permissions from the platform they are integrated into.
The Role of Regulators
Regulators are starting to address the issue of third-party agents by requiring enterprises to inventory and oversee their AI systems. This is a step towards better security and compliance, but it is not a complete solution.
Conclusion
The third-party agent problem is a significant challenge for enterprise security. It requires continuous monitoring and understanding of agents' behavior, as well as a shift in how security teams approach the management of third-party agents.
Key points
- Security struggles with third-party AI agents that are not easily monitored or controlled.
- Agents often operate without a clear identity, making it difficult for security to track them.
- Agents can reach across multiple systems and data stores, creating a larger attack surface.
- Agents can perform actions that are not aligned with their intended use.
- Regulators are starting to address the issue of third-party agents, but more needs to be done.
As security teams learn more about third-party agents, they can develop better tools to monitor and control them. This will make it easier to identify and address security issues.
If security teams don't keep up with the growth of third-party agents, they might miss important security issues. This could lead to data breaches or other security problems.
