Autonomous agents powered by Large Language Models (LLMs) are susceptible to malfunction amplification attacks. These attacks exploit the inherent instability within these agents by inducing repetitive or irrelevant actions through methods such as prompt injection and adversarial perturbations. This results in agent malfunction and task failure. The subtle nature of these attacks makes them difficult to detect using standard LLM safety mechanisms, as they do not rely on overtly harmful actions. Mitigation steps: **For AI Developers:** * Implement comprehensive input validation and sanitization protocols to prevent prompt injection vulnerabilities. * Integrate advanced anomaly detection systems that surpass basic policy violation assessments. * Incorporate systems to identify and halt repetitive or irrelevant actions that surpass established thresholds. **For Model Trainers/Fine-tuners:** * Continuously evaluate agent behavior and responses to diverse inputs to uncover potential vulnerabilities. * Utilize a variety of Large Language Models (LLMs) for agent cores to mitigate the effects of vulnerabilities specific to individual models.