Resource-Abuse Detection and Economic DoS Prevention

Determine how to distinguish legitimate high-volume use from adversarial resource abuse, detect prompts that induce excessive reasoning or tool calls, and control economic denial-of-service risks in cloud-based LLM services.

Background

Unbounded consumption attacks exploit the computational cost of inference by using long prompts, repeated requests, excessive reasoning, or chains of tool calls. Although rate limiting, resource caps, and anomaly detection provide practical defenses, the paper states that these controls do not reliably distinguish legitimate intensive use from abuse. Autonomous agents make the problem more difficult by chaining multiple model interactions, APIs, and external tools, potentially creating substantial operational costs and service degradation.

References

How to properly calibrate models is an open problem.

— Instrumental Monitor Evasion Emerges Under Ordinary Task Pressure  (2609.30217 - Schmotz et al., 24 Sep 2026) in Conclusion

Although rate limiting, resource capping, and anomaly detection provide practical defenses, key challenges remain. It is still difficult to distinguish legitimate high-volume use from adversarial resource abuse, detect adversarial prompts that trigger excessive reasoning or tool calls, and control economic denial-of-service risks in cloud-based LLM services. Autonomous agents further amplify this risk by chaining model interactions, API calls, and external tools.

— Shifting from Injection to Interaction: Rethinking Web Security in the Age of LLMs and Beyond  (2609.03999 - Singh et al., 3 Sep 2026) in Section 5, subsection LLM_10: Unbounded Consumption, Open Research Challenges box

Slurm already supports mechanisms such as per-QoS limits and bounded job-array concurrency, but deciding where these constraints should be enforced, either in the agent, its harness, or the scheduler, remains an open question.

— Towards Efficient HPC Systems for Agents: Challenges and Opportunities  (2609.38723 - Zheng et al., 30 Sep 2026) in Section Compute, subsection “Opportunities: Trial-Aware, Hardware-Aware, and Event-Driven Scheduling”