The Problem
As it did with the advent of the internet, criminals are now using AI to support their activities. With its help they can create more convincing deepfakes, voice clones, synthetic identities, official documents and fake profiles, which are making it increasingly difficult to authenticate information. We have come across a number of examples including our clients’ being provided with fake bank statements, websites and company documents.
The Risk
The risk to companies is greater than ever, as the speed, type and number of scam and fraudulent approaches has increased dramatically. Fraudsters, with the help of AI and other sources, can now create apparently legitimate individuals, businesses and relationships, building complex structures, profiles and backgrounds to convince companies of their authenticity. Sophisticated fraudsters seek to establish credibility and relationships before attempting the actual fraud.
The Investigative Challenge
How do you distinguish genuine information from manufactured information?
AI can be a very useful part of a verification process, but an AI detector’s “real” or “fake” answer should never be treated as proof. The evidence supports this caution: NIST currently reports that deepfake-detection systems can lose 45–50% of their performance when moving from academic testing to operational deployment.[1]
NIST is the US National Institute of Standards and Technology, part of the US Department of Commerce. As part of its remit it runs the NIST AI Challenge Problems, which conduct community evaluations to advance safe and trustworthy AI.[ Meanwhile, the technology that generates convincing fakes continues to improve, making the detectors’ task harder. (https://www.nist.gov/about-nist; https://ai-challenges.nist.gov/ )
Verification
It is incumbent on the party concerned to check the veracity of any information provided as part of a business deal, relationship or similar.
Source provenance, corroboration, human intelligence and traditional investigative methodology therefore remain important.
For example, when checking the veracity of a document, AI can assist in looking for suspicious characteristics (such as spelling, grammar, fonts and logos), but you should not rely on AI alone to authenticate its appearance. An investigator will independently verify the document by contacting the organisation it came from, using contact details found independently (i.e. not from the document provided). This is just one example; a later newsletter will examine other verification methods in more detail.
Conclusion
AI does not eliminate the need for investigators. In fact it may make professional investigation more important. In an environment where information can be manufactured, almost instantly, the value increasingly lies not simply in finding information but in determining whether it can be trusted.
If you are unsure about the authenticity of a counterparty, document or approach, please get in touch.
