What happened
Fake airline customer service numbers are reaching travelers through documents hosted on trusted websites, social media posts and AI-generated summaries. In research published on October 21, 2026, Check Point Exposure Management analysts Ben Neeman and Ruty Davidson describe a campaign that mass-produces fraudulent customer support PDFs and posts for more than 20 travel brands, including airlines, travel agencies and transportation companies. The researchers say the campaign was still active when they published.
The reported method is simple. Templated "how to contact" guides carry a brand name and a phone number that does not belong to that brand. The files are uploaded through public forms and user-content features on third-party sites, including university webforms, job application portals and government document-submission systems, where they receive public URLs and can be indexed. Check Point shows examples served from a Utah state government portal and indexed on the website of the Moldovan Consulate General in Chicago. Those organizations hosted the files. The report does not describe them as the impersonators.
The same numbers then appeared elsewhere. Check Point reports fraudulent numbers in Instagram advertisements and Facebook pages, and in Facebook search summaries labeled "Summarized by Meta AI." In one example, searches related to Booking.com customer service returned an AI-generated answer containing a fraudulent contact number. The researchers describe the numbers as consistent with callback-fraud operations, in which callers may be pressured to pay fees or disclose booking and payment information.
Several facts remain unknown from the public report. It does not publish a count of documents, calls, victims or losses, name the operators or list every impersonated brand. It does not say whether Meta or the hosting sites have removed the material. The payment and data requests are described as what callers "may" face, so this article treats them as the researchers' assessment of the scheme's purpose rather than as measured outcomes. None of the impersonated brands is reported to have suffered a breach of its own systems.
Where Outtake could have helped
The following analysis is hypothetical. It applies Outtake's current published capabilities to the external stages Check Point describes. It does not imply that any affected brand uses or lacks Outtake, that Outtake took part in this research, or that earlier action would have prevented any call or loss.
Treat the phone number as the thread
Outtake Brand Protection describes monitoring domains, social platforms, app stores and paid advertising for unauthorized use of a brand, and mapping the operation behind individual fakes. In this campaign, the brand name changes between documents while the same numbers recur across a hosted PDF, an Instagram ad and a social post. That repetition is the useful signal. An investigation that starts with one fraudulent ad or post could support linking other public assets that advertise the same number, then packaging that evidence for each platform involved.
The value of connecting them is practical. A team that removes one post without recognizing the shared number may see the same contact detail reappear somewhere else. A connected view could help a brand decide which assets to report together and which hosts to notify first.
Connect customer reports to the campaign
Many brands learn about fake support numbers when a customer complains after calling one. Outtake Intake describes evaluating customer and employee reports as they arrive, extracting indicators and connecting isolated complaints into a single campaign. Applied here, a traveler's report that names the number they called and where they found it could become the starting point for finding the documents and posts carrying that number. Customer reports could also show which surfaces travelers actually used, including an AI summary the brand would otherwise not see.
Support action with the right provider
The Outtake platform describes threat graphing, evidence packaging, automated reporting and platform escalation. This campaign spreads across very different providers: a university or government site that accepted an upload, an advertising platform, a social network and an AI summary feature. Each has its own reporting route and its own definition of abuse. Evidence that shows the brand claim, the false number, the hosting location and the connection to other assets could support requests to each of them. Removal decisions remain with those providers and site owners.
Outtake's published materials do not specifically describe monitoring the answers that AI assistants give. The relevant role here is upstream, in the public documents, posts and ads those answers can draw on. Earlier action against those sources could reduce the material available for repetition, but this article does not claim it would remove a number from an AI-generated answer.
What the October research adds
A related campaign at a different entry point
Check Point's September 16 report described scammers replying to public complaints on X, Facebook and Instagram while posing as airline support, then moving customers to WhatsApp. The October research describes a campaign aimed at an earlier moment, the search for a support contact before the customer has posted anything. Check Point calls the new campaign related. It does not say the same operators run both, so this article treats them as separate campaigns with a shared target.
Evidence of increasing automation
The October report says production became more automated over seven weeks. Researchers observed shared templates, files generated seconds apart and incrementing counters, and describe a progression from Google Docs exports to scripted generation and a custom automation application. The report does not give the start date of that seven-week period. For defenders, the practical consequence is volume. When a new document costs almost nothing to produce, removing individual files without addressing the repeated numbers, accounts and upload paths gives the operator little reason to stop.
Why file scanners did not flag it
Researchers rechecked confirmed campaign files with between 61 and 64 antivirus engines per scan, and none detected them. That result reflects what the engines are built to find. The PDFs contain no exploit, macro or payload. The harm is the false claim that a number belongs to a brand. Detection therefore depends on comparing a document's claimed affiliation and contact details with the brand's real channels, not on inspecting the file for code.
Why trusted hosting changes the investigation
The host is not the impersonator
Check Point identified upload paths associated with Drupal webforms, WordPress form plugins, Optimizely and custom government portals. Because many of these forms legitimately accept PDFs, a fraudulent document can pass ordinary file validation and end up at a public address on a trusted domain. An investigation needs to separate three parties: the brand being impersonated, the organization whose site hosts the file, and the operator who uploaded it. Treating the host as the attacker would misdirect the report and could slow removal.
Independent reporting shows how large one host's exposure can become. Más Colombia reported on October 18 that a subdomain of Bogotá's culture secretariat was receiving PDFs presented as airline and travel-agency contact guides, with more than 5,000 indexed in seven days and the oldest located entry dated June 18, 2026. The outlet traced sampled entries to a webform upload directory and said it could not confirm how the files arrived. That reporting does not establish a link to the operators in Check Point's research. It does show the same pattern of templated guides on a public-sector upload path.
Short-lived files still leave evidence
Más Colombia found that most files stopped opening within hours, while new ones kept appearing. A file that is gone on a second check may still have been indexed, summarized or copied elsewhere. Teams should record the search result, the hosting URL, the time observed and the number displayed when they first see a document, rather than relying on a later visit. A missing file does not prove the number has stopped circulating.
AI answers repeat what they retrieve
This is not prompt injection
Nothing in Check Point's report describes instructions hidden for an AI model. The researchers assess that question-style titles, repeated phone numbers, keyword-heavy copy and placement on authoritative domains may raise the chance that a false number is retrieved and summarized. Our explainer on phishing sites and AI agents covers prompt injection, where content tries to redirect an agent. This campaign needs no such instruction. It only needs the false information to look like a credible answer.
A pattern documented before
Other researchers have reported similar activity. Aurascape described poisoned PDFs on government, university and WordPress sites feeding fake airline numbers into AI answers in December 2025. In September 2026, Dark Reading reported Vigilance Security research on a campaign it calls Dark Sourcery, which found fabricated support details attributed to at least 374 companies. These are separate disclosures and do not confirm the details of Check Point's campaign. Together they show that fake support numbers in AI-generated answers are a recurring technique rather than a one-off error.
What a travel brand could observe
Check Point recommends that travel brands monitor search results, social platforms, file repositories and AI-generated answers for fraudulent support information, and make verified channels easy to find. Several signals follow from the reported mechanics:
- Documents titled as guides for reaching your support team that sit on domains unrelated to your business.
- One phone number attributed to your brand and to other brands in different documents.
- Social ads or posts that promote a support or refund number not listed on your official channels.
- Customer complaints that mention calling a number found through search or an AI summary.
- Support contacts in AI-generated answers that do not match your published numbers.
None of these signals proves fraud alone. A third-party travel agency can legitimately list its own number. The question is whether a document or post presents a number as the brand's own without authorization. Related guidance on paid search abuse covers how sponsored placements can borrow a brand's identity in a similar way.
Making the real number easier to find
Check Point's guidance to brands pairs monitoring with making verified support channels easy to find. The reasoning follows from the campaign design. A traveler who needs help during a cancellation or a missed connection is under time pressure and often starts with a search rather than the airline's app. If the brand's own support page is hard to reach, or does not state its contact numbers in plain text, a templated guide on a trusted domain can look like the more direct answer.
Más Colombia reported that KLM advises customers not to search for its phone number online and to use its official website and app instead. Guidance of that kind works best when customers see it before they need help, for example in booking confirmations, check-in messages and disruption notices. Brands can also state plainly which contact channels they use, so customers have a reference point when a number or a caller's request seems wrong.
These steps do not remove fraudulent documents. They reduce the chance that a traveler trusts one, and they give customer service teams a consistent way to recognize reports that point to an impersonation campaign rather than a single confused caller.
Limits of this analysis
The campaign facts above come mainly from one primary source. Check Point's findings have been widely republished, but republished copies are not independent confirmation, and the independent reporting cited here covers related observations rather than the same dataset. The report does not quantify calls, payments or victims, and it does not name operators.
Any external intervention would depend on public visibility of the documents, posts and ads, and on the cooperation of hosting organizations, advertising platforms, social networks and AI providers. Short file lifetimes, private conversations after a call and payment through gift cards or cryptocurrency all sit outside what public monitoring can observe. A brand's own customer communications and the security of the sites that accept uploads remain essential and separate controls.
Follow a fake support number through the campaign
Choose a view, then select a stage. The campaign flow follows the five stages Check Point reported. The surface view groups the places a brand team could observe the same number. Possible Outtake interventions are hypothetical. This educational section collects no data and displays no fraudulent numbers.
Campaign flow reported by Check Point
1. Manufacture
Templates turn one number into many brand guides
Reported evidence
Check Point reports keyword-heavy guides generated in bulk from shared templates, with brand names and phone numbers swapped between documents. Researchers saw files generated seconds apart and a move from Google Docs exports to scripted generation over seven weeks.
Possible external intervention
Repeated template wording and recurring numbers could support grouping documents that impersonate different brands into one investigation, rather than handling each file as a separate case.
Control boundary
The production workflow itself is not publicly visible. Only documents that are published and discoverable can be observed.
2. Seed
Public upload features place files on trusted domains
Reported evidence
Documents were uploaded through university webforms, job portals, government submission systems and other public forms that expose files at public URLs, according to Check Point. Its examples include a Utah state government portal.
Possible external intervention
Evidence identifying the hosting organization, the file location and the false brand claim could support a notification to the site owner, who controls the upload path.
Control boundary
Only the site owner can close or secure an upload form. Hosting organizations are not the impersonators, and their response times vary.
3. Index
Search and AI summaries ingest the content
Reported evidence
Search engines, AI answer engines and platform summaries ingest the content. The numbers appeared in Facebook search summaries labeled "Summarized by Meta AI," per Check Point.
Possible external intervention
Reducing the public sources that carry a false number could reduce what remains available to be retrieved. Customer reports could reveal which AI summaries travelers saw.
Control boundary
Outtake's published materials do not describe monitoring AI assistant answers. Changes to an AI-generated answer depend on its provider.
4. Capture
A traveler searches for help and calls
Reported evidence
A traveler dealing with a cancellation, missed flight or delayed refund finds the number and calls, as described in Check Point's five-stage model. The report does not count calls.
Possible external intervention
A report from a traveler who called could be connected to the same number in public documents and ads, turning one complaint into evidence of a wider campaign.
Control boundary
The call itself happens on the phone network, outside public web visibility. Detection at this stage relies on what customers choose to report.
5. Monetize
Callers may face fees or data requests
Reported evidence
Check Point assesses that callers may be pressured to pay fraudulent fees or disclose booking and payment information, with possible payment through gift cards, wire transfers, peer-to-peer payments or cryptocurrency.
Possible external intervention
Evidence connecting the number to the campaign could help a brand warn customers through official channels and share details with relevant payment or platform partners.
Control boundary
This stage is the researchers' assessment, not measured victim data. Recovering payments and securing bookings depend on the brand, banks and payment providers.
Where a brand team could see it
Hosted document
A PDF claims to be your support guide
Reported evidence
Check Point found fraudulent PDFs on public-sector, institutional and commercial sites. Separately, Más Colombia counted more than 5,000 indexed travel guides on one Bogotá government subdomain in seven days, without linking them to the same operators.
Possible external intervention
Comparing the document's claimed affiliation and number with the brand's verified channels could support a documented report to the host.
Control boundary
Files may disappear within hours, so evidence has to be captured when first seen. A missing file does not prove the number stopped circulating.
Social ad or post
An ad promotes the same number
Reported evidence
Check Point reports fraudulent numbers in Instagram advertisements and Facebook pages. One pair of numbers appeared across a hosted PDF, an Instagram ad, a creator post and a state portal PDF.
Possible external intervention
Outtake Brand Protection describes monitoring social platforms and paid advertising. A shared number could support reporting related ads and posts together.
Control boundary
Platforms decide whether an ad or page violates their policies. Removing one ad does not address documents hosted elsewhere.
AI-generated summary
An AI summary presents the number as support
Reported evidence
Searches related to Booking.com customer service returned a Meta AI summary containing a fraudulent number, according to Check Point. Earlier research by Aurascape and Vigilance Security described similar answers on other AI systems.
Possible external intervention
Investigating the public sources an answer may draw on could support reports to those sources and give the AI provider documented evidence.
Control boundary
Outtake's published capabilities do not include monitoring or editing AI answers. Only the provider controls what its system returns.
Customer complaint
A customer reports the number they called
Reported evidence
Check Point's September research and its October report both describe travelers seeking help being diverted to fraudulent support. The October report does not publish complaint data.
Possible external intervention
Outtake Intake describes extracting indicators from complaints and connecting related reports. A reported number could lead to the documents and ads carrying it.
Control boundary
Complaints arrive after contact has occurred. They show only what customers notice and choose to report.
Frequently asked questions
Was any airline or travel company breached in this campaign?
No breach of the impersonated brands' systems is reported. Check Point describes third parties publishing false contact details under those brands' names on other organizations' websites and on social platforms. The brands' own infrastructure is not described as compromised.
Why did antivirus engines not detect the fraudulent PDFs?
The files contain no malicious code. Check Point rechecked confirmed campaign files with 61 to 64 antivirus engines per scan, and none flagged them. The harm is the false claim that a phone number belongs to a brand, which requires comparing the content with the brand's verified channels.
Is this the same campaign as Dark Sourcery?
That is not established. Dark Reading reported Vigilance Security's Dark Sourcery research in September 2026, and Check Point published separately in October. Both describe fake support details reaching AI-generated answers, but neither report links its findings to the other's operators or data.
Can a brand remove a fake number from an AI-generated answer?
Only the AI provider controls its answers. A brand can report inaccurate answers through the provider's channels and address the public sources that carry the number, such as hosted documents, posts and ads. Outtake Brand Protection covers those public sources, and Intake covers inbound reports. Neither is described as editing AI answers directly.
What should organizations that accept public uploads check?
Check Point recommends authentication and moderation controls, keeping uploaded files out of crawlable directories, reviewing temporary upload paths and auditing publicly accessible storage. These controls belong to the site owner and are separate from a travel brand's own monitoring.
Sources
- Check Point, How Fraudulent Airline Support Numbers Exploit Search Results and AI-Generated Answers, October 21, 2026.
- Check Point, Scammers Are Watching Airline Complaints and Posing as Customer Support, September 16, 2026.
- Más Colombia, Un subdominio de la Secretaría de Cultura de Bogotá alojó más de 5.000 guías falsas para viajeros, October 18, 2026 (Spanish).
- Dark Reading, Attackers Manipulate AI in Mass Disinformation, Phishing Campaign, September 23, 2026.
- Aurascape, When AI Recommends Scammers, December 2025.
- Outtake platform, Outtake Brand Protection and Outtake Intake.