Trust, but Verify—with AI
By Dennis D. McDonald — and ChatGPT
The problem
I have been receiving unsolicited messages through LinkedIn from people who say they are recruiting advisers or consultants for artificial intelligence companies. The details vary, but the pattern is remarkably consistent. The company is often a startup on the West Coast or in Asia. The work is remote or part time. The proposed role involves AI strategy, business development, project management, or advising senior executives. The message also mentions enough from my LinkedIn profile to suggest that someone has done some homework and knows something about my professional interests. I’ve described the process here and, half-seriously, I’m wondering if these inquiries are being generated in Singapore.
The offers can sound interesting. They are tailored to the recipient, technically plausible, and written in competent business English. They also tend to share one peculiar requirement: the recruiter wants to move the conversation immediately from LinkedIn to Signal, WhatsApp, or Telegram.
I prefer email for an initial business contact. My email address is readily available on both my LinkedIn profile and my public website. Email also gives me something I can investigate: a domain, an organization, an address, and a record of the exchange.
When I tell these recruiters that I will continue by email, preferably through a corporate account, most “recruiters” disappear.
The “French” Connection
One of my earliest experiences involved a person who said she was French. She was patient, conversational, and considerably more convincing than the usual spammer. She supplied a photograph and continued the exchange long enough to make me wonder whether the inquiry might be genuine. But she insisted on WhatsApp, resisted email, and never gave me enough information to verify who she was. I could not even persuade her to provide her last name or an email address. I finally cut off the conversation and resolved never to use WhatsApp again.
The Recurring LinkedIn Pattern
The LinkedIn messages started up a few months ago and have continued, sometimes at the rate of one or two a week. They were similar enough to suggest a reusable process. Each offered some variation on an advisory position with an AI startup. Each drew details from my background in consulting, proposals, market research, technology management, and strategic planning. Each quickly proposed moving to a private messaging application. I also know how easily tools like ChatGPT can imitate a conversational, personalized style. LinkedIn itself lists immediate pressure to move off the platform as a warning sign of a job scam.
The “ABEJA” Inquiry
The most interesting example arrived recently. A LinkedIn contact said she was recruiting for ABEJA USA, Inc., described as the North American operation of ABEJA, Inc., a legitimate Japanese AI company. I again requested email. This time I received it, but the message came from a Gmail account rather than an ABEJA address.
The email was technically impressive. It referred to enterprise AI, agentic AI, physical AI, business architecture, transformation strategies, proofs of concept, and scaled implementation. The proposed position fit my background well enough to be genuinely interesting.
Some research changed the picture. I located a public GitHub warning describing an earlier and very similar campaign by the same named “recruiter” claiming to represent another legitimate AI company. According to that account, the approach began through Gmail, used polished but vague technical language, attempted to move the target to WhatsApp, and eventually supplied a fraudulent lookalike corporate domain.
The ABEJA material contained an additional twist. ABEJA really does advertise a “DX and AI Consultant Business Architect” position—in Japan. ABEJA's public corporate profile did not identify the claimed American operation or a specific recruiter. The inquiry borrowed enough real details to make the claimed relationship believable.
What the Scammer May Want
The ultimate purpose remains uncertain because I stopped these exchanges before any fraud became explicit. The supposed position may merely be a way to establish a relationship with someone who has professional credentials, financial resources, and useful contacts. After one or more interviews, the sender might request identity documents for onboarding, banking information for payroll, payment for equipment or background checks, installation of interview software, participation in a cryptocurrency investment, or introductions to other professionals. Another possibility is the extraction of free consulting advice under the guise of evaluating a candidate.
Using AI as an Investigative Assistant
Based on this experience and the supporting research, ChatGPT suggested in one of our “conversations” the following verification steps:
Preserve the original LinkedIn message, profile URL, email headers, phone number, and any documents or links.
Search the sender's full name with the company name, email address, phone number, and quoted phrases from the message.
Confirm that the company and any claimed subsidiary exist through sources independent of the sender.
Locate the position on the company's official careers site. Compare location, duties, reporting relationship, compensation, and employment status.
Check whether the sender appears on the official website, in credible professional history, or among people clearly associated with the company's verified LinkedIn page.
Examine the email domain carefully. A free account is not proof of fraud, but a senior internal recruiter should ordinarily be able to use a corporate account. Newly registered or misspelled domains require special caution.
Contact the company through an address, telephone number, or web form located independently. Do not use contact information supplied by the questionable recruiter. (I have contacted Abeja using its contact form and am awaiting a resopnse.)
Refuse early requests for identity documents, financial information, payments, software installation, or confidential information until the relationship has been verified.
Report suspicious accounts and conversations to LinkedIn. A public post may warn colleagues, but it is not the same as sending a report into LinkedIn's review process.
A tool could turn these checks into an evidence-based legitimacy score. It should show why the score was assigned rather than simply label the inquiry safe or fraudulent. Also, LinkedIn could perform some checks that an outside service cannot.
Using AI to Counter AI?
There is some irony in asking an AI system to help determine whether someone else is using AI to deceive. This connects with a broader issue I discussed while reading the National Academies report on advances in artificial intelligence and cybersecurity. Attackers need only one approach to succeed, while defenders must watch many possible points of failure. That National Academies report argues that AI tools themselves will have to become part of cybersecurity defense because purely human processes may not keep pace with AI-supported attacks.
Recruiting scams are a modest but revealing example of the same problem. AI can generate polished approaches, personalize them with public information, and revise them when a target resists. These messages can be produced quickly, in large numbers, and at low cost.
Ronald Reagan repeatedly advised, “Trust, but verify.” That is especially good advice when a job opportunity seems too good to be true!
Author’s Note
This article was developed collaboratively by Dennis D. McDonald and ChatGPT in a series of online conversations in September 2026. McDonald supplied the experiences, argument, judgments, and editorial direction. ChatGPT assisted with research, organization, source comparison, and preparation of an initial draft. McDonald reviewed, revised, and substantially shortened that draft to what is published here. Note: This work was done using a paid ChatGPT account. A free account can perform many of the same investigative steps, although its access to advanced models, file analysis, research tools, and memory across conversations may be more limited. Also, Flaticon is the source of the image at the head of this article.
