LinkedIn has developed a new AI image detector that claims to be able to detect 99.6% of fake profile pictures, with a false positive rate of 1%. According to anecdotal reports, their new detector actually works.

Fake LinkedIn profiles

There are many reasons why people create fake LinkedIn profiles.

For some members of the affiliate search marketing community, one reason for fake profiles is to assume that Google trusts a site if the article’s authors have links to a LinkedIn profile in their author bios.

This idea stems from Google’s recommendation that content should have what is known as EEAT, experience, expertise, authority and trustworthiness.

For others, the motivation is to offer their website visitors a more trustworthy website.

This is not to condone these practices, I advise against them.

This is only to explain that the practice is taking place and why it is taking place.

The advent of the ability to create profile pictures using AI has made it easier to create fake profiles, exponentially exacerbating an already major problem.

Reports of fake LinkedIn profiles released in 2022 state that LinkedIn detected and removed 21 million fake accounts in the first half of 2022.

Anecdotal evidence from an affiliate marketer who deployed fake LinkedIn profiles confirms that LinkedIn’s AI image detector significantly improved its ability to detect fake accounts.

According to LinkedIn:

“We are constantly working to improve and increase the effectiveness of our anti-abuse safeguards to protect the experiences of our members and customers. And as part of our ongoing work, we’re collaborating with academia to stay ahead of new types of fake account abuses that take advantage of rapidly evolving technologies like generative AI.”

Fake accounts are difficult to detect

LinkedIn is constantly updating its systems to detect various types of unwanted activity, such as fake profiles, account takeovers, and content policy violations.

The advent of AI-generated images has made it nearly impossible to spot fake images if you don’t know what to look for.

LinkedIn identifies “artifacts” that are the hallmark of fake AI profile pictures.

Most people don’t know how to spot AI images, so it’s easy to mistake a fake account for a real one.

LinkedIn shared:

“With the advent of AI-generated synthetic media and text-to-image generated media, fake profiles have become increasingly sophisticated.

And we’ve found that most members are generally unable to visually distinguish real faces from synthetic faces…”

How LinkedIn collects AI-generated content

A characteristic of man-made images is that they all share similar patterns, which LinkedIn calls structural differences.

Real images have no common structural components.

LinkedIn shared an example of a composite of 400 artificial images and 400 real images.

The composition of the fake images shows that the areas around the eyes and nose tend to be very similar.

The composition of the real images has nothing in common with the other images, so the composition is blurred.

Screenshot of image from LinkedIn

The results of their research are impressive.

LinkedIn Stocks:

“True Positive Rate (TPR) is the percentage of synthetic photos that are correctly classified as synthetic.

False Positive Rate (FPR) is the percentage of real photos that are incorrectly classified as synthetic.

Our approach is able to detect 99.6% (TPR) of StyleGAN, StyleGAN2 and StyleGAN3 synthetic faces, while only 1% (FPR) of genuine LinkedIn profile photos are incorrectly classified as synthetic.

For the benchmark results in our research, we chose an FPR target of 1% because it is important for real-world applications in a large professional network that AI-generated image recognition models capture most of the synthetic images, albeit rarely a real image classified as synthetic.”

How effective is the LinkedIn AI Detector in the real world?

The affiliate marketer with the fake LinkedIn profiles shared that LinkedIn was able to intercept 100% of their fake LinkedIn profiles.

They shared their experiences with me:

“As an affiliate marketer, creating LinkedIn profiles for my fake persona was a great way to give my writers credibility.

This was particularly helpful for HARO link building, as reporters tend to link more often to websites featuring people with a LinkedIn profile.

In the last few months, 90% of my profiles have been banned from LinkedIn.

Unfortunately, I now have to find another way to give my authors credibility and make them look serious.”

LinkedIn continues to improve its ability to detect fake profiles. The ability to create a fake profile is now even harder.

Read the original announcement:

New approaches to recognize AI-generated profile photos

Featured image from Shutterstock/Meilun