Microsoft and Salesforce Fix Data Leak Vulnerabilities in AI Agent Platforms

Microsoft and Salesforce have issued patches addressing security flaws in their AI agent platforms that could have allowed unauthorized access to sensitive data. The vulnerabilities stemmed from how these AI agents handle data permissions and context boundaries, potentially enabling attackers to extract information the agents had access to through crafted interactions.

The fixes highlight a growing attack surface as enterprises rush to deploy agentic AI systems that operate with broad access to internal data stores, CRMs, and communication tools. Security researchers have been warning that AI agents inherit the permissions of the systems they connect to, creating opportunities for data exfiltration if prompt injection or context manipulation techniques bypass intended guardrails.

The patches underscore the need for organizations to treat AI agent deployments with the same rigor as any privileged service account - applying least-privilege access, monitoring agent behavior, and validating that platform vendors are actively securing the agent-to-data pipeline.