AI assistants are rapidly moving beyond answering questions to accessing data, using tools, and taking actions on behalf of users. As those capabilities expand, so do the potential security risks.
Security researchers recently demonstrated how vulnerabilities in a major retailer’s AI shopping assistant could be chained together to manipulate the system and ultimately reach capabilities it was never intended to expose. For MSPs helping clients adopt AI, the incident highlights several important security considerations.
1. Prompt Injection Can Become a Serious Threat
Researchers demonstrated how malicious instructions embedded in external content could influence the AI assistant’s behavior. As AI systems become more connected to business tools and data, prompt injection can potentially lead to much greater consequences.
MSP Action: Evaluate how client AI systems interact with external content and prevent untrusted information from automatically triggering sensitive actions.
2. AI Agents Need Clearly Defined Permissions
An AI assistant should only have access to the systems, tools, and information necessary to perform its intended job. Excessive permissions can increase the potential damage if an agent is manipulated.
MSP Action: Review the permissions, APIs, tools, and data available to client AI agents and apply least-privilege principles wherever possible.
3. Basic AI Guardrails May Not Be Enough
Controls designed to keep an AI assistant focused on its intended purpose don’t necessarily protect the entire system. MSPs need to consider the integrations and backend capabilities surrounding the AI model as well.
MSP Action: Assess the complete AI workflow, including external data retrieval, APIs, tool calls, and backend actions—not just prompts and responses.
4. AI Systems Need Ongoing Security Testing
AI agents can change as models, integrations, and capabilities evolve. Security assessments shouldn’t end when an AI application is initially deployed.
MSP Action: Include AI systems in ongoing vulnerability assessments, security reviews, and testing programs as part of the client’s broader technology environment.
5. AI Governance Is Becoming a Security Priority
AI governance isn’t just about deciding which tools employees can use. Businesses increasingly need policies governing what AI systems can access, what actions they can perform, and how those activities are monitored.
MSP Action: Help clients establish AI governance policies covering permissions, data access, monitoring, security testing, and accountability.
Helping Clients Secure AI Adoption
The retailer incident demonstrates how security risks can grow as AI moves from generating answers to taking actions. For MSPs, the opportunity is to help clients adopt these technologies while building appropriate security and governance around them.
As AI agents become more capable and connected, providers that understand their risks can play an increasingly important role in helping clients automate securely.