Why did United Airlines block AI agents from booking?

Why did United Airlines block AI agent access?

Recently, United Airlines made a drastic technical decision to block access from various third-party artificial intelligence agents to their booking platforms. To the average user, this might seem like a setback in the convenience of having an AI that plans your entire trip. However, from the perspective of engineering and mission-critical systems infrastructure, the move resolves serious performance and security issues. The real problem the airline is trying to mitigate is the use of autonomous agents that ignore acceptable use policies, performing constant scans that degrade the performance of their booking servers.

In practice, this means that when an AI agent scans flight inventory every second to find the best price, it acts as an unintentional denial-of-service (DDoS) attack. These automation pipelines, if not well-managed, consume bandwidth and processing power that should be reserved for real customers making purchases. Furthermore, United seeks to prevent dynamic pricing data from being extracted and used to feed unfair competition platforms.

The mechanics behind the block: API protection

Under the hood, United utilized advanced Web Application Firewall (WAF) solutions and machine learning-based behavioral analysis to identify non-human access patterns. These systems track browsing behavior, request frequency, and the validity of credentials. When an AI bot attempts to simulate a real user (so-called 'scraping'), it often fails by not completing the full authentication handshake or by presenting HTTP header signatures that diverge from legitimate browsers.

The technical implementation of protection involves blocking IP addresses based on reputation and requiring session tokens that expire rapidly. Unlike simple bots, modern AI agents have reasoning capabilities, but they still depend on API endpoints to fetch prices. By restricting this access, the airline forces developers to seek formal agreements, such as using official APIs, where latency and transaction costs are controlled through Service Level Agreements (SLAs)—a topic we often discuss when evaluating decision automation at scale.

AI vs. Travel Aggregators: A comparative impact

Currently, the travel AI market is divided between agents that use controlled browsers (like Playwright or Selenium) and those that consume structured data. Traditional aggregators have decades-long partnerships with airlines, ensuring that inventory is displayed accurately. Newer AIs, often based on frameworks like OpenAI Swarm, try to bypass these intermediaries to offer a personalized experience, but without guarantees that seating or baggage information is synchronized.

The critical disadvantage of third-party AIs compared to official tools is latency. An agent that must 'pretend' to be a human to navigate the site takes much longer to finalize a booking than a direct API integration. While the AI takes seconds to fill out fields, the API resolves the transaction in milliseconds. This creates a quality gap where the AI user faces a higher risk of finding a price that no longer exists at the time of purchase (the infamous 'stale data' problem).

The verdict for developers: How to evolve?

For developers working with automation and AI, the United Airlines block serves as a wake-up call: the era of unrestricted access to airline data is ending. The trend is for companies to create developer 'sandboxes' where access is monitored and paid. For those building agents, the path forward is to focus on integrating official APIs (such as NDC - New Distribution Capability) rather than just attempting to 'hack' the user interface. Attempting to bypass these protections is not only legally risky but also makes the application unstable, as firewall rules change almost daily.

If you are developing a tool that depends on real-time data, consider that your software's robustness depends on the stability of the data source. Instead of building a 'web scraper' that can be blocked at any moment, it is worth investing in partnerships to access aggregator APIs. Legitimate automation, based on standard protocols and secure authentication, is the only kind that survives in the long term within complex ecosystems like commercial aviation.