AI Security Atlas › All entries › LLM Application & Output Security
Denial of Wallet & Resource Exhaustion
Floods of requests, oversized inputs, or deliberately expensive prompts drive up inference cost or take the service down.
Open in the interactive atlas →Impact
Unplanned cloud bills, degraded service for legitimate users, or a full outage.
Why this severity
Impact Medium × Likelihood Medium = Medium
Impact medium: financial and availability loss, usually recoverable. Likelihood medium: cost harvesting is documented with stolen credentials (LLMjacking); public-endpoint floods are less documented.
Impact Medium: Limited or recoverable harm: cost, availability, reputational damage, or exposure confined to one session or user.
Likelihood Medium: Publicly demonstrated against real products or in peer-reviewed research, or seen in the wild only with a prior foothold such as stolen credentials.
Method: OWASP Risk Rating Methodology
Evidence
- Incident: LLM Jacking (MITRE ATLAS case study AML.CS0030, 2024)
Related
Exploits
- No Per-Request Resource Limits Vulnerability · Medium severity
Mitigated by
- Resource Budgets, Timeouts & Graceful Degradation Guardrail · Fixes medium risks
Remediation
- Cap input size, output tokens, and execution time per request.
- Apply per-user and per-tenant quotas with budget alerts.
- Degrade gracefully (shorter answers, queueing) instead of failing outright.
How to verify
- Load-test with oversized and adversarially expensive prompts.
- Confirm budget alerts fire before spend exceeds the monthly cap.
Framework mapping
OWASP Top 10 for LLM Applications 2025
- LLM10:2025 Unbounded Consumption
MITRE ATLAS
NIST AI RMF 1.0
- MEASURE 2.7 AI system security and resilience are evaluated and documented