Getting the Right Security in Place for Agentic AI
This post highlights the growing importance of Agentic AI in modern businesses, leveraging Generative AI for autonomous decision-making and task execution with minimal human oversight.
The Value of Agentic AI
Agentic AI automates workflows across various organizational functions, such as:
- Triage cybersecurity threats
- Personalize marketing materials
- Handle returns
- Manage inventory
By incorporating mission logic, these AI systems can learn from their outcomes and continuously improve, which is why it’s projected that by 2027, half of all enterprises utilizing Generative AI will deploy AI agents.
Emerging Security and Governance Challenges
With the advantages of Agentic AI come significant security and governance challenges, as noted in the report “The Automated Enterprise: Agentic AI and the New Security Imperative”:
Access Control and Security
Organizations typically rely on access control lists to safeguard their data. However, as AI agents operate across multiple systems, new methods for controlling these agents and their permissions are essential.
Hallucinations and Cascading Failures
The risk of hallucinations or inaccurate information arises when Generative AI relies on approximations for output. Errors in communication between AI agents can result in a series of cascading failures. Utilizing technologies like Vertex AI Search can ground models in enterprise data, ensuring outputs are factual and relevant.
Skills and Experience Gaps
The development and deployment of enterprise-grade Agentic AI systems demand highly skilled personnel. The current shortage of knowledgeable employees poses security challenges, emphasizing the need for a solid security groundwork.
ROI and Navigating the Unknown
While the outlook for return on investment (ROI) in AI is improving due to decreasing costs and advances like model distillation, some leaders remain cautious about the unpredictable behavior of autonomous agents in critical environments.
A Security Framework for Agentic AI
To ensure the security of Agentic AI, a structured methodology is recommended:
- A governance framework: Align AI initiatives with organizational strategies. Frameworks like Deloitte’s Trustworthy AI™ Framework provide governance and risk controls to align AI with enterprise strategies and regulatory requirements.
- Human oversight: As AI scales rapidly, implementing a human-in-the-loop review process at key checkpoints is necessary to identify risks early.
- Data reliability: To prevent bias, using trusted enterprise data is crucial for enhancing decision-making and reducing AI bias.
AI for business is advancing swiftly, driven by more efficient models and innovations. Organizations that successfully implement the right controls and security frameworks today are more likely to trust and utilize autonomous agents in complex, high-priority scenarios in the future.
