- What did United Airlines actually do, and why does it matter for your company?
- What exactly does United Airlines' policy say?
- Why is robots.txt alone no longer enough?
- What does an Italian company without an AI policy really risk?
- Three Real-World Scenarios: What Changes for Different Italian Companies?
- What tools exist to protect company data from AI agents?
- Why is this being discussed now: is the AI market in Italy accelerating?
- How to build an AI access policy: the concrete steps
- What mistakes are made most often when tackling this topic?
- How do I know if my AI policy is working?
- So, is it worth building an AI policy for your company?
- How we see it at SHM
- Frequently Asked Questions
- United Airlines has banned AI agents (autonomous software that books or collects data) from accessing its site without written permission, separating this rule from robots.txt.
- 30% of AI bots already ignore robots.txt files: technical rules alone are no longer enough, you need explicit legal terms.
- The Italian Data Protection Authority (Garante Privacy) has put AI among its 2026 inspection priorities: for Italian companies, the risk of having no policy is real.
- Building a policy for AI access to your data isn't just a technical matter: it's a strategic choice about who can use your information and how.
A company updates its terms of service and, from November 11, 2025, any AI software that accesses its site without written authorization is in breach. It's not science fiction: it's what United Airlines did, and the model is spreading fast.
What did United Airlines actually do, and why does it matter for your company?
United Airlines has updated its terms of service to prohibit any autonomous AI software from accessing its site or app without prior written permission. The move isn't just about airlines: it sets a model that any company with proprietary data online can follow, and maybe should.
On November 11, 2025, United Airlines updated its terms of service and drew a clear line: traditional search engines can index the site's public pages, but any AI agent — that is, software that acts autonomously to search, book or buy — must first get written permission from the company.
CEO Scott Kirby explained the reasoning plainly: "I don't want to create another middleman." In other words, the company doesn't want third parties using its fare, availability and pricing data to train AI models or to operate as unauthorized booking agents.
If you run an e-commerce store, a B2B portal or any site with proprietary data, the question is immediate: is my data already accessible to AI software I haven't authorized?
| Your situation | Worth acting? | why |
|---|---|---|
| You have an e-commerce store with real-time prices and availability | Yes | Your data is an asset: a competitor's AI agent can use it to make comparisons or train models |
| You run a B2B portal with restricted price lists | Yes | Unauthorized access can breach commercial agreements and confidentiality regulations |
| You have a corporate website with generic public content | Partly | The risk is lower, but your content can be used to train AI models without consent |
| You already use an updated robots.txt | It's not enough | 30% of AI bots ignore it: you need explicit legal terms and additional technical protections |
| You're an SME with no sensitive data online | No, for now | Low priority, but it's worth monitoring how Italian regulation evolves |
What exactly does United Airlines' policy say?
United's terms define an 'Automated Agent' as any software — including AI assistants, bots, browser extensions and features built into browsers — that acts autonomously on the site or app. None of these can operate without written permission, and using the data to train AI models is explicitly prohibited.
The definition is deliberately broad. An Automated Agent includes any software that, with full or partial autonomy, searches, books, buys or manages items on the site, or extracts or copies data. This category covers AI assistants, bots, web scrapers (programs that collect data automatically), browser extensions and even the AI features built into modern browsers.
The rule applies even if a person supervises the agent's actions. United reserves the right to block any agent at any time, regardless of who operates it.
| What's allowed | What's prohibited |
|---|---|
| Search engines indexing public pages | AI agents that book or purchase without written permission |
| Assistive technologies for people with disabilities | Browser extensions with autonomous AI features |
| Standard browsers used by a person | Using data to train, fine-tune or evaluate AI models |
| Researching and planning trips with AI (without accessing the site) | Travel agencies using third-party AI agents without authorization |
| Bookings made before November 11, 2025 | New bookings through unauthorized AI agents |
One important technical detail: these rules don't depend on the robots.txt file — the document that traditionally tells bots what they can and can't do on a site. United chose to rely on legal terms, not just technical instructions. As of November 14, 2025, United's robots.txt didn't name any specific AI crawler.
This isn't an isolated case. Amazon blocked Meta's Muse AI agent based on its own conditions of use. eBay added "buy-for-me agents" to its list of automated tools that require explicit permission.
Why is robots.txt alone no longer enough?
Robots.txt files are technical instructions that bots are supposed to follow, but they have no legal force. The data shows a growing share of AI bots systematically ignore them.
Robots.txt was created in the 1990s to tell search engines which pages not to index. It worked well when bots were few and well-behaved. Today the landscape has changed.
According to TollBit network data, in the fourth quarter of 2025 about 30% of AI bots ignored robots.txt permissions, up from 13% in the first quarter of the same year. In just one year, the share more than doubled.
Websites responded quickly. According to TechnologyChecker.io, blocking responses to AI bots rose from 3.63% of requests in the second quarter of 2025 to 8.56% in the second quarter of 2026. But technical blocking isn't enough without a legal basis to back it up.
To learn more about how autonomous AI agents work and the risks they bring, the article on autonomous AI models and corporate compliance offers a complete overview.
What does an Italian company without an AI policy really risk?
The risk isn't just theoretical: the Italian data protection authority (Garante Privacy) has made AI one of its inspection priorities for 2026, and the data shows AI agents violate terms of service in almost all of the cases analyzed.
A study published on arXiv in November 2026 — the ScrapeBench benchmark — analyzed the leading autonomous AI agents across 1,626 sites that explicitly prohibit data collection. The legal compliance rate was below 5%. The agents violated the terms mainly through misinterpretation, not malicious intent. But from a legal standpoint, the distinction matters little.
On the Italian regulatory front, the Garante per la protezione dei dati personali has put the use of AI among its 2026 inspection priorities (Provvedimento 797 of December 30, 2025), with a specific focus on data breaches, big data and corporate AI. In 2025 it already carried out 130 inspections and handled 4,288 complaints with a response.
| Type of risk | Description | Who's most exposed |
|---|---|---|
| Unauthorized use of data | An AI agent collects prices, catalog or customer data without permission | E-commerce, B2B portals, SaaS platforms |
| Training of competing models | Your content is used to train a competitor's AI | Companies with proprietary content, professional firms |
| Regulatory violation | The lack of an AI policy may amount to a violation of the GDPR or Garante guidelines | All companies with personal data online |
| Unauthorized purchases or changes | An AI agent makes transactions on a customer's behalf without human oversight | E-commerce, subscription services, travel |
| Loss of control over proprietary data | Price list or availability data used by third parties for automated comparisons | Manufacturers, distributors, agencies |
Three Real-World Scenarios: What Changes for Different Italian Companies?
The problem looks different depending on the industry. Here are three illustrative examples to show what's at stake.
Illustrative examples: the numbers and situations are meant to show the reasoning, they are not real customer data.
| Company type | Data at risk | Scenario without a policy | Scenario with a policy |
|---|---|---|---|
| Clothing e-commerce store (500 SKUs) | Real-time prices, availability and discounts | A comparison site's AI agent collects prices every hour and uses them to steer customers toward competitors | The terms prohibit automated access; the technical block is backed by a legal basis |
| Tax consulting firm | Training content, guides, proprietary FAQs | The content is used to train a competitor's chatbot without consent | An explicit policy prohibits using the content for AI training; the site includes a clause in its terms |
| Manufacturing company with a B2B catalog | Confidential price lists, warehouse availability, technical specifications | A customer's AI agent accesses the portal and downloads the price lists to automatically compare them with other suppliers | Portal access requires authentication, and the terms prohibit unauthorized automated use |
What tools exist to protect company data from AI agents?
Effective protection combines three layers: legal terms, technical rules and active monitoring. None of the three is enough on its own.
| Tool | What it does | Limits | Recommended for |
|---|---|---|---|
| Updated robots.txt | Tells bots what not to index | 30% of AI bots ignore it; no legal force | Minimum baseline for all sites |
| Explicit terms of service | Legally defines what AI agents are not allowed to do | Needs periodic updates and legal advice | Companies with valuable proprietary data |
| Authentication and rate limiting | Blocks automated access to restricted pages | Doesn't protect public pages | B2B portals, e-commerce with confidential price lists |
| WAF (Web Application Firewall) | Filters abnormal traffic and blocks known bots | Needs configuration and continuous updates | High-traffic sites or sites with sensitive data |
| Bot traffic monitoring | Identifies suspicious automated access patterns | Needs dedicated tools and analytical skills | Companies that want to know who is accessing their data |
The United model shows that the most robust approach is to combine clear legal terms with technical protections. Legal terms give you a basis to act in case of a violation; technical protections reduce the risk that a violation happens. As we saw when looking at AI search agents and their ambiguities, the problem is often not bad faith but the lack of clear rules that agents can interpret correctly.
Why is this being discussed now: is the AI market in Italy accelerating?
Yes. The Italian AI market reached €1.8 billion in 2025, growing 50% in a year. But adoption among SMEs is still very low, which means many companies are exposed without adequate tools.
According to the Artificial Intelligence Observatory of Politecnico di Milano, the Italian AI market reached €1.8 billion in 2025, up 50% from 2024. Adoption, however, is very uneven.
The gap is stark: 71% of large Italian companies have launched at least one AI project, compared with just 8% of SMEs. And according to the 2026 data from the Osservatorio Innovazione Digitale nelle PMI, 76% of Italian SMEs have neither invested nor plan to invest in AI, and only 7% have started structured training programs.
This means many SMEs find themselves in a paradoxical position: their data is already accessible to third-party AI agents, but they don't yet have the tools to see it or protect themselves. The issue of AI agents amplifying incorrect data shows clearly how a lack of governance can create cascading problems.
How to build an AI access policy: the concrete steps
You don't have to start from scratch: the United model offers a practical reference point. The steps are sequential and require input from whoever manages the site, your legal team and your IT team.
- 1 · IT / Web TeamMap your exposed dataIdentify which pages contain proprietary data: prices, availability, restricted content, customer data. These are the areas to protect first.
- 2 · Legal / ConsultantUpdate your terms of serviceAdd an explicit definition of "AI automated agent" and prohibit access without written permission. Also specify the bans on using data for AI training.
- 3 · IT / Web TeamUpdate your robots.txtAdd specific rules for the main known AI crawlers. It isn't enough on its own, but it reduces unauthorized traffic and strengthens your legal position.
- 4 · IT / Security TeamImplement technical protectionsSet up rate limiting (a cap on the number of requests per IP), authentication for restricted areas and a WAF to filter abnormal traffic.
- 5 · Marketing / IT TeamMonitor bot trafficTurn on automated traffic monitoring. Look for unusual patterns: many requests in a short time, visits to catalog or price pages at odd hours.
- 6 · ManagementDefine the authorization processDecide who can grant written permission to third-party AI agents and under what conditions. For now, United has chosen not to authorize anyone.
What mistakes are made most often when tackling this topic?
The most common mistake is thinking the problem only affects large companies, or that an updated robots.txt is enough. It isn't.
Working with companies across different industries, we've seen the same mistakes repeat. Here they are, in order of frequency.
- Leaving everything to robots.txt. It's a starting point, not the solution. Without explicit legal terms, you have no tools to act in case of a violation.
- Thinking the problem only affects those with sensitive data. Editorial content, technical guides and product catalogs are assets too, and they can be used to train competing AI models.
- Not updating your terms of service for years. Most Italian terms of service don't mention AI agents. They're documents written before these tools existed.
- Ignoring your supply contracts with agencies. United also updated its contract with travel agencies. If you work with partners who access your systems, they must be bound by the policy too.
- Not monitoring traffic. Without data, you don't know if you're already suffering unauthorized access. Monitoring is the first step to understanding the scale of the problem.

How do I know if my AI policy is working?
An effective policy is measured on three levels: reduced anomalous bot traffic, no detected violations, and alignment between technical rules and legal terms.
| Indicator | How it is measured | Positive signal |
|---|---|---|
| Bot traffic on the site | Analytics + server logs | Fewer abnormal requests after the terms update |
| 403 responses to AI crawlers | WAF or server logs | More block responses to known bots |
| robots.txt / legal terms consistency | Periodic review of documents | Robots.txt and terms prohibit the same categories of agents |
| Permission requests received | Email / dedicated form | Formal requests coming from third parties |
| Reports of unauthorized use | Web monitoring / alerts | No mentions of your data in unauthorized contexts |
It's worth remembering that a policy isn't a static document. United has said it will update its FAQs as the AI agent landscape evolves. The same approach applies to any company: a twice-yearly review of your terms and technical protections is the reasonable minimum in a field that changes this fast. If you manage complex data flows, tools like those described in the article on Google Data Manager for audit and targeting with AI agents can help you better understand what's moving through your systems.

So, is it worth building an AI policy for your company?
Worth it if you have proprietary data online (prices, catalog, restricted content, customer data) and don't want third-party AI software collecting or using it without consent. Even an SME with an active e-commerce store or a B2B portal has every reason to act.
Somewhere in between: your site has generic public content, but you're growing or introducing restricted areas. It's worth updating your terms of service now, before the problem becomes urgent.
Not yet if your site is purely corporate, with no significant proprietary data and no restricted areas. Keep an eye on how Italian regulation evolves, but it's not an immediate priority.
How we see it at SHM
For us, the issue of AI agents and access to company data is a perfect example of how innovation should be analyzed before it's chased — or endured. United Airlines didn't block AI: it chose to govern it, setting clear rules on who can access its data and under what conditions. It's exactly the approach we recommend to our clients.
We believe AI should be integrated to free companies from routine, not to create new unmanaged risks. And that works both ways: when a company adopts AI tools, but also when it needs to protect itself from third-party AI agents.
We've worked with companies that managed catalogs with thousands of SKUs across multiple markets, where control over who accesses the data and how they use it was a concrete operational priority. In those contexts, data governance wasn't an abstract legal topic: it was the difference between a reliable catalog and one exposed to outside manipulation.
If you're evaluating how to structure AI governance in your company, we can help you start with strategic AI integration, a review of your systems with a data processing and business intelligence approach, or AI training for your team. You'll find all our in-depth articles on the topic in the AI and process automation section.
The point isn't to block AI. It's to set the rules of the game yourself, before others set them for you.
Frequently Asked Questions
It depends on what it does and how your terms of service are written. If your terms explicitly prohibit automated access and the agent collects personal data, it may amount to a GDPR violation. If it collects proprietary commercial data, it may amount to a copyright or unfair competition violation. Without explicit terms, it's much harder to take action.
Both, but with different roles. Robots.txt reduces unauthorized bot traffic at a technical level, but it has no legal force. Terms of service create the legal basis for acting in case of a violation. United Airlines chose to rely mainly on legal terms, keeping robots.txt as a complementary tool.
Analyze your server logs and traffic with advanced analytics tools. Look for unusual patterns: many requests in a short time from the same IP, visits to catalog or price pages at unusual hours, unknown or suspicious user agents (software identifiers). Many WAFs (web application firewalls) offer specific bot traffic reports.
Yes, and it's exactly the model United chose: written permission case by case. You can set out an authorization procedure in your terms, specify which uses are allowed and which aren't, and revoke permission at any time. It's the most flexible solution for anyone who wants to work with selected AI partners without losing control.
No, the policy covers third-party AI agents that access your site from the outside. The chatbots you install on your own site are tools you control. That said, if your chatbot pulls data from other sites to answer users, it's worth checking that it complies with those sites' terms.
The cost depends on the complexity of your site and on your legal counsel. For an SME with a standard site, a targeted update of your existing terms takes just a few hours of legal work. That's far less than the risk of having no legal basis in case of a violation, especially now that the Garante Privacy has made AI one of its inspection priorities for 2026.
The real question isn't whether to block AI agents, but who decides the rules for accessing your data: you, or anyone with a bot fast enough.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.