When attackers make your AI unavailable
AI agents can be overwhelmed by malicious traffic or expensive requests, making them unavailable for legitimate use and potentially incurring significant costs.
Denial of Service attacks aim to make your AI agent unavailable by overwhelming it with requests or exploiting resource-intensive operations. For AI agents, DoS is particularly concerning because:
- LLM API calls are expensive—attackers can drain your budget - Complex prompts can consume significant compute - Queue buildup can delay legitimate requests indefinitely - Auto-scaling can be exploited to increase costs
Unlike traditional web services, AI agents have asymmetric resource usage—a small request can trigger expensive processing. This makes them attractive DoS targets.
Overwhelming the service with sheer request volume.
Crafting requests that consume disproportionate resources.
Sending prompts designed to maximize token usage and processing time.
Using the AI to generate large outputs that consume bandwidth.
Filling async processing queues so legitimate work never executes.
Making enough API calls to exhaust monthly quotas or budgets.
An exposed AI assistant became the target of a DoS attack:
1. Attackers discovered the endpoint had no rate limiting 2. They scripted requests with prompts designed to maximize token usage 3. Each prompt asked for detailed, lengthy responses 4. Within hours, the monthly API budget was exhausted 5. Legitimate users couldn't use the service 6. The company received a surprise $15,000 bill from their LLM provider
The attack cost cents to execute but thousands of dollars in damage.
When you self-host your OpenClaw, you're responsible for addressing these risks:
Clawctl includes built-in protection against dos attacks:
Configurable rate limits prevent request flooding. Per-key and global limits available.
Enterprise-grade DDoS mitigation at the network edge. Volumetric attacks never reach your agent.
Set hard spending limits. The system stops processing before exceeding your budget.
Malformed or suspiciously complex requests are rejected before processing.
Scaling has caps to prevent runaway costs during attack attempts.
Whether you use Clawctl or not, follow these best practices:
Clawctl includes enterprise-grade protection against this threat and many others. Deploy your OpenClaw securely in 60 seconds.