Executive Summary
Mandiant's testing found prompt injection remains a primary attack vector in enterprise AI, alongside weak file permissions and access controls, and recorded an agent that ran up a $50,000 cloud bill.
What Happened
Mandiant’s testing found prompt injection remains a primary attack vector in enterprise AI deployments, alongside weak file permissions and poor access controls. The report also highlighted an AI agent that generated a large cloud bill during testing, showing how agentic systems can create direct financial risk.
Vulnerability / Attack Technique
Prompt injection; runaway agent resource consumption
Technical Mechanism
Tests found injection and permission weaknesses; one agent consumed cloud resources unchecked until the bill reached $50,000.
AI Security Category
LLM10:2025 Unbounded Consumption
Relevant Frameworks
RealGround Analysis
Agents create direct financial risk as well as data risk. RealGround analysis: budgets and quotas are security controls for agents.
Recommended Controls
- Set spend limits and quotas for agent cloud and API usage
- Alert on unusual agent resource consumption
- Fix file permissions agents can reach
Related RealGround Service
Secure AI Agent Build · or start with the free AI Agent Security Scanner.
Sources
Help Net Security: https://www.helpnetsecurity.com/2026/09/16/google-mandiant-enterprise-ai-security-risks-report/
Summary and analysis written by RealGround from the source above; last updated 2026-10-17.