DoS Attacks on AI: Technical Defense Guide | Quiz

Why: Coordinated API resource exhaustion requires activating circuit breakers to halt new requests and fail over to healthy instances while investigating the coordinated attack pattern across multiple accounts. Context: Each account is within limits individually making per-account rate limiting ineffective. Circuit breakers at 95% GPU utilization for more than 5 seconds protect against this coordinated abuse. Remember: Coordinated attacks need infrastructure-level circuit breakers not per-account limits.

Why: This pattern indicates a sponge example or token bomb attack – the attacker is sending complex requests that consume massive resources while staying under traditional rate limits. Context: Traditional rate limiting by request count would not catch this because the number of requests is normal. The attack exploits computational asymmetry where one complex request costs 50x a normal request. Remember: Normal request count but abnormal cost equals sponge/token bomb attack.

Why: Server-side output caps enforce maximum output tokens at the infrastructure level regardless of what clients request – this single control blocks the majority of token bomb attacks. Context: Never trust client-requested output limits. Attackers will always request maximum output to maximize resource consumption. Server-side enforcement is essential because it cannot be bypassed. Remember: Never trust client limits – enforce output caps at infrastructure level.

Why: Computational asymmetry means the attacker-defender cost ratio heavily favors attackers – crafting an expensive prompt takes seconds but processing it takes minutes of GPU time worth significant money. Context: A 10-word prompt can trigger a 4000-word response with complex reasoning. The attacker invested virtually nothing while the defender spent real compute resources. Remember: Attackers spend seconds – defenders spend dollars in GPU time.

Why: Traditional protection uses request-count rate limiting but one complex AI request can consume 1000x the resources of a simple request – making request counts meaningless for AI systems. Context: CDNs and IP-based rate limiting cannot detect that a complex prompt costs vastly more than a simple one. AI DoS requires token-aware rate limiting and input complexity analysis. Remember: Request counts are meaningless – AI needs token-aware rate limiting.

Why: Sponge examples are inputs specifically designed to maximize processing time by triggering extended chain-of-thought reasoning and recursive patterns and maximum-length generation. Context: These attacks exploit model architecture weaknesses. A single request can tie up GPU resources for minutes instead of seconds while legitimate users queue up or time out. Remember: Sponge attacks maximize processing time – like ordering the most elaborate dish on the menu.

Why: The three attack types are sponge examples/token bombs (50-500x cost multiplier) and compute-heavy prompts (10-200x cost) and model degradation via poisoning (permanent damage). Context: Each type has different mechanisms and detection difficulty. Sponge attacks maximize output tokens while compute-heavy attacks force deep reasoning and model degradation poisons training data. Remember: Three types – sponge/token bombs and compute-heavy and model degradation.