There is just one problem: parts of it may be wrong.
Artificial intelligence can make intelligence research faster. It can assist with document review, translation, comparison and the organisation of large volumes of data. But an AI-generated intelligence report is not reliable merely because it is fluent, detailed or accompanied by citations. Every material claim still needs verification.
Why plausible errors are dangerous
Large language models can produce false statements that sound confident, sometimes called hallucinations. They can also overlook context, merge information about different people or turn a tentative association into an apparently established relationship.
The UK's National Cyber Security Centre has publicly warned that AI systems can present incorrect statements as facts and can be influenced by misleading input.
In ordinary research, an error is inconvenient. In a sensitive private-client or corporate investigation, it may affect a person's reputation, a transaction, an employment decision or a client's safety.
An invented litigation reference or incorrectly attributed company directorship could change how an advisor assesses risk. Equally serious is an omission that conceals a genuine uncertainty behind a polished executive summary.
What should a family office verify?
If an AI-assisted report is used to evaluate a business associate, advisor or potential transaction, the decision-maker should be able to answer several questions.
Where did each material claim come from? Is the original source accessible? Does the source actually support the conclusion? Are dates and identities correct? Which assertions were corroborated independently? What information is missing or disputed?
Those questions may sound basic. They are also where automated research often needs the most human oversight.
A source link is not proof. A linked page may have changed, be unreliable or fail to say what the report claims. And a database hit may concern the wrong person.
AI is a tool, not an accountable investigator
The right comparison isn't human research versus machine research. It is a controlled intelligence process versus an uncontrolled one.
Good analysts can use technology to speed up suitable tasks while preserving source traceability, data protection, verification and clear analytical ownership. Sensitive client information should only be processed through systems approved for the engagement and applicable confidentiality requirements.
For important findings, the investigator must remain answerable for how the conclusion was reached. The person commissioning the work should know which judgements are evidence-based, which are provisional and where additional enquiries are justified.
The FRYA position
An attractive PDF does not protect a principal. A correct assessment might.
FRYA approaches investigations by defining the question, evaluating available material and using appropriate specialist capabilities to establish what can responsibly be concluded.
Technology can assist. It cannot replace context, corroboration or professional judgement. Especially when the decision affects a person's access, assets or safety.