The Authentication Conundrum: The Ground is Shifting

Given the proliferation of high-quality AI generated and altered real images, detecting images that lack authenticity is getting more challenging. It was not always this way. Until relatively recently, AI generated imagery involving people as subjects was easily detectable because of obvious body morphology anomalies such as extra limbs, fingers, toes, and improbable body positioning. Similarly, some images involving outdoor scenes, buildings, vehicles, and other objects could be identified as AI generated because of obvious content repetitions and other incongruencies. Increasingly however, establishing image authenticity requires a level of expertise not present in untrained viewers.

Arguably, image authentication has reached what might be called an existential crisis. It is often easier to prove that an image is not authentic by finding even one provable anomaly than it is to prove that an image is authentic. An image can be shown to be inauthentic if alterations are detected or AI generated content can be identified within an image. Obvious anomalies like extra limbs and other low-hanging visual cues may not require expert assistance but most such authentication challenges will require expert intervention as such assessments are beyond the reasonable capabilities of counsel and the court. There is thus scope for determining if an image is not authentic providing the appropriate expert resources are marshalled.

Conversely, it may prove to be just as challenging to proactively establish that an image is authentic. There was a time when it was judicially acceptable for a person with some level of qualifications to say that an image is authentic because that person could not find any evidence of alteration or AI involvement. Those days have now passed. The value of the statement ‘I did not find any evidence of alteration or generative AI…’ must now be evaluated in a broader context. Is the declarant sufficiently qualified to offer an opinion on authentication such that a court should accept it as true? Did the declarant use the appropriate tools and methodology to make that assessment? Did the declarant miss a clue that points to image alteration or AI involvement? A blind person saying that they don’t see something would naturally be accorded less weight than a sighted person making such a statement. It follows that a declaration that an image is authentic will carry much more weight if it is made by a highly qualified person using appropriate tools and methodology and who has thoroughly examined the images. A bold declaration of authenticity by someone who is not so qualified and did not use appropriate tools and methodology, even though sincere, is worth far less. It follows that assessments of authenticity should be made by the correct people using the correct tools and methodology and who thoroughly assess the images.

Authentication is a threshold issue. This means that it is a precondition to evaluating image for reliability, interpretive value, and proof purposes. Therefore, competent resources must be front-loaded into cases wherein authenticity is a potential issue. Note that not all cases will be authentication puzzles ripe for solving. When the photographer is known and the original media is available for technical evaluation, authentication will be easier to establish using active forgery detection techniques. Ideally, investigators and other parties of interest would have access to the original media and the photographer, but increasingly such access is more constrained and often impossible. This is especially so when images are uploaded to social media and other open source sites in anonymity and where access to the original media is non-existent. This problem exists in domestic cases and even more so in international criminal law cases.

It is essential that the assessment of image authenticity be given a prominent role in the evaluation of images. It is not sufficient for a technician or analyst to state ‘I did not find any evidence of alteration or generative AI…’ without being able to give that statement the objective backing it requires. When we learned math in grade school, we were required to show how we got the answer – the methodology was integral to evaluating whether the answer was meaningful. Thus, technicians and analysts must also be able to show the math when they assess image authenticity. By showing the competent use of technology, forensic methodology, and sound practices, a conclusory statement about the presence or lack of authenticity will have more weight. It will also allow the parties to make evaluative decisions about image content, its meaning, and its value for proof and truth-seeking purposes.