An AI research connection is a lead, not a finding

A company sees an engineer leave for a competitor. Soon afterwards, the competitor publishes work in a similar area. That sequence may justify questions. It does not, by itself, establish that information was taken. The commercial concern is real; the explanation still needs investigating.

The ATIU work examines collaboration, movement, compensation and the structure of AI research teams. These are useful starting points because technical expertise develops through people working together. The author list of Attention Is All You Need is a documented example of a shared contribution. Later company-founding records show how some of those relationships extend beyond the original team. Co-authorship establishes collaboration on that paper; it does not establish a continuing private relationship or access to a present employer’s information.

For a client, the useful question is specific. Are we assessing a prospective hire, a research partnership, a supplier or a suspected disclosure? A map that is useful for recruitment diligence may be insufficient for investigating an information leak. Before collecting more data, decide what finding would change the commercial decision.

An insider enquiry requires a different layer of evidence: the information at issue, who had access, the relevant dates, authorised disclosures and any records of transfer. Investigators can compare those records with interviews and the public timeline. A prior research relationship may explain why two people know each other. It cannot substitute for evidence of what they did.

Supplier exposure needs the same discipline. Establish ownership, control, contractual access and technical dependencies. A person’s nationality does not demonstrate divided loyalty, and a company’s location does not tell us what information it can access. The enquiry should focus on the actual relationship and its consequences.

Cato & Veil brings research mapping, corporate investigation, financial analysis and technical examination into one brief. The output should explain which connections are documented, what they mean for the decision and which questions remain open. That is how a visually compelling network becomes useful intelligence.

Cato & Veil analysis, informed by the ATIU research corpus and the public records linked below.