AI Dictionary › Fondamenti AI

LLM Denial of Service

Denial of Service su LLM

LLM denial of service is a category of attack in which an attacker seeks to degrade or interrupt the availability of a service based on a language model, not necessarily by stealing data or manipulating output, but simply by exhausting its computational or economic resources. Unlike classic denial of service against web servers, which relies on saturating bandwidth or network connections, attacks against LLMs exploit the specific characteristics of inference computation.

Definition

How it works

Techniques include sending prompts crafted to maximize processing time or the length of the generated output, potentially exploiting longer reasoning cycles in models with extended reasoning capabilities; sending a high volume of legitimate but useless requests to saturate available capacity and drive up compute costs or response times for other users; or prompts that, exploiting specific model weaknesses, induce it into repetitive loops that consume tokens without producing useful output.

Applications

It is an economic risk as well as a technical one for anyone offering a pay-as-you-go AI service: since language model APIs are typically billed based on tokens processed, an attack of this kind can translate into significant direct costs on top of service degradation for legitimate users. Main defenses include rate limits, maximum limits on input and output length, and monitoring of anomalous usage patterns.

History & etymology

The concept adapts to the language model domain a well-known cyberattack category dating back to the 1990s in the context of networks and web servers; the specific application to LLMs emerged with the spread of paid commercial APIs starting in 2022-2023.

Related terms

• Rate Limiting
• Max Tokens
• Prompt Firewall
• Misuse Prevention
• Throughput

More in Fondamenti AI

• LLM (Large Language Model)
• Token
• Context Window
• Temperature
• AI Hallucination
• Fine-tuning
• RAG (Retrieval Augmented Generation)
• Embedding

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.