A security team is threat-modeling an LLM feature using STRIDE. They identify the threat: "An attacker repeatedly sends maximum-length prompts to exhaust the model serving infrastructure." Which STRIDE category is this?

Why A is correct

Repeatedly sending maximum-length prompts to exhaust inference compute, memory, or API quotas maps directly to Denial of Service in STRIDE. LLMs are particularly vulnerable to DoS via token amplification: input tokens that generate very long outputs disproportionately consume GPU compute. Mitigations include token-based rate limiting, maximum input length enforcement, and output length caps. Spoofing involves impersonating principals. Tampering modifies data. Information Disclosure exposes sensitive data.