LLM10: Unbounded Consumption
Denial of Wallet Challenge
Explore how uncontrolled LLM usage can lead to excessive resource consumption and costs. Complete challenges to understand different aspects of unbounded consumption vulnerabilities.
Understanding Unbounded Consumption
What is Unbounded Consumption?
Unbounded Consumption occurs when LLM applications allow excessive and uncontrolled inferences, leading to resource exhaustion, financial losses, and service disruption. This vulnerability exploits the high computational demands and cost-per-use model of LLM services.
Attack Vectors
- Input Flooding: Overwhelming with variable-length inputs
- Resource Drain: Complex, CPU-intensive queries
- Chain Reactions: Self-perpetuating prompts
- Cost Exploitation: Maximizing token usage
Impact
- Service degradation and outages
- Unsustainable operational costs
- Resource exhaustion
- Denial of service to legitimate users
Example Attack Prompts
Here are some example prompts that demonstrate different types of unbounded consumption attacks:
Token Flooding
Write a comprehensive guide about artificial intelligence, including its history, current applications, future potential, and ethical considerations. Make it extremely detailed with many examples and use cases.
Resource Intensive
Create a complex sorting algorithm in Python that handles multiple data types, includes detailed comments explaining each step, and analyze its time complexity with different input sizes.
Chain Reaction
Generate a story outline. Then, for each plot point, expand it into a detailed scene. For each scene, describe the characters, setting, and dialogue in extreme detail.
Cost Accumulation
Analyze this code base: [paste a large codebase]. Explain each function in detail, suggest improvements, and provide example usage for each component.
Token Flood
Make the API process more than 1000 tokens in a single request
Cost Explorer
Accumulate $0.01 in API costs through efficient token usage
Chain Reaction
Create a self-perpetuating prompt that generates increasingly longer responses
Resource Drain
Trigger high CPU usage through complex processing requests
OpenAI API Configuration
Prevention Strategies
Technical Controls
- Implement strict rate limiting
- Set token usage quotas
- Monitor resource consumption
- Use timeouts and circuit breakers
Best Practices
- Validate and sanitize inputs
- Implement cost monitoring
- Set up usage alerts
- Use graceful degradation