SIGNAL//DESK
AI securitysrc: Roost FA glossary v2

Cost overrun

AI calls cost money. Loops, retries, or long contexts can spiral fast — Project Vend, June 2025. In your FA practice, this looks like an automated agent stuck in a retry-loop, burning your monthly API budget in under an hour.

In AI security, a cost overrun occurs when an automated agent enters an infinite retry-loop or processes unnecessarily large contexts, causing a rapid, unexpected depletion of your compute budget as defined by Project Vend.

A cost overrun is the rapid, unexpected depletion of compute budgets caused by unbounded recursive loops, excessive token generation, or inefficient prompt chaining in production AI pipelines, which can result in total exhaustion of allocated API funds within minutes.

evolution

  1. 2022-11 · history
    LLM API Commercialization

    The release of ChatGPT introduced token-based billing models, creating the fundamental economic risk of runaway API consumption.

  2. 2023-05 · history
    Agentic Loop Discovery

    Early autonomous agent frameworks like AutoGPT demonstrated how recursive task loops could trigger massive, unintended API expenditure.

  3. 2024-02 · history
    Prompt Injection Economic Attacks

    Security researchers identified 'denial-of-wallet' attacks where malicious prompts force models into infinite loops to exhaust enterprise budgets.

  4. 2025-06 · history
    Project Vend

    The formalization of cost-overrun as a critical AI security vulnerability, highlighting the need for automated budget circuit breakers in agentic systems.


← all terms