AI deepfakes are spreading online: why is removing them so difficult?

An image shared online can quickly travel beyond its original audience, appearing on websites, social media accounts and messaging platforms without the knowledge or consent of the person depicted. Artificial intelligence has made it easier to create manipulated images and videos, adding to concerns about privacy, identity misuse and the difficulty of controlling content once it spreads.

Removing such material can be more complicated than reporting a single post. Copies may be downloaded, edited, reposted or shared across different platforms, each with its own reporting procedures and rules for handling complaints. Even when one version is taken down, other copies may remain accessible.

Why removing an image is so difficult

The first step in addressing image misuse is often to locate the content and report it to the website hosting it. However, finding one instance does not necessarily reveal every copy circulating online. A manipulated image may appear under different account names, be uploaded with a different caption or be altered slightly to make it harder to locate. Some copies may also be shared in private groups or on websites that are not easily indexed by search engines.

Removal procedures vary between platforms. Some require users to submit specific forms or evidence, while others may ask for identity verification or additional information before reviewing a complaint.

The process can therefore involve several stages: locating the image, documenting the misuse, submitting a report and checking whether the content has been removed. If new copies appear, the process may have to begin again.

Removing an image from one website also does not necessarily remove it from search results, other websites or devices where it has already been saved. The outcome depends on the platform, the type of content and the action taken by the relevant service.

How AI is changing image misuse

AI tools can generate or manipulate images and videos in ways that make it increasingly difficult for viewers to distinguish authentic material from fabricated content. Deepfakes may use a person's likeness in situations they never participated in, including sexualised or otherwise harmful scenarios.

The consequences can extend beyond embarrassment. People targeted by manipulated content may face harassment, damage to their professional reputation, privacy violations and emotional distress.

The risks are not limited to public figures. Ordinary social media users, professionals and content creators can also have their photographs copied or manipulated without permission.

At the same time, identifying a potentially manipulated image is not always straightforward. An unusual-looking image is not necessarily a deepfake, and a resemblance between two images does not by itself establish that one has been altered. Detection tools can help identify material for further assessment, but their results need to be interpreted carefully.

Can technology help locate and report fake images?

Technology services are increasingly combining image monitoring, potential manipulation detection and removal-request management into a single process.

One example is Strike by Synctric, a service that describes itself as helping users monitor online image misuse and manage requests to remove content. According to the product information supplied by the company, its features include recurring checks of selected face-search sources, monitoring across online platforms and AI-based assessment of potentially manipulated images.

The service also says that human review is used before supported removal requests are submitted. This distinction matters because automated systems can flag potential matches or suspicious content, but those results do not automatically establish that an image is fake or that a particular platform must remove it.

Bringing monitoring and complaint management together may help users organise evidence and track reports. However, the effectiveness of such a process depends on the sources being monitored, the accuracy of the detection tools and the responses of the websites hosting the content.

The company's founders, Aqib Muhammed and Ketan Ramshawala, have framed the service around using technology not only to identify potential misuse but also to help people respond to it. That is the company's stated approach; its actual effectiveness would need to be assessed through independently verified results.

Why detection does not guarantee removal

Identifying a potentially harmful image and getting it removed are two different challenges.

A detection system may find a copy on a monitored website, but it cannot necessarily access every online platform or private channel. A removal request may also be rejected, delayed or require additional documentation, depending on the platform's rules and the circumstances of the complaint.

Even after a successful takedown, copies may persist elsewhere. This means no monitoring service can guarantee that every instance will be found or permanently erased from the internet.

Users dealing with image misuse can start by recording the relevant URLs, preserving appropriate evidence and reporting the content through the platform's official channels. Where intimate images or other potentially illegal material are involved, they may also need support from relevant authorities or legal professionals.

People should avoid resharing harmful material while trying to draw attention to it, as doing so can increase its reach.

Protecting digital identity requires more than AI

As image-generation tools become more accessible, the challenge of protecting digital identity is likely to involve several complementary measures. Monitoring tools can help locate potential copies, detection systems can flag material for review, and organised reporting processes can help users follow up on complaints.

Privacy safeguards also matter. Services handling sensitive images and evidence need clear rules governing access, retention and deletion. Strike by Synctric's product information, for example, describes access controls for case evidence and deletion of retained files after 30 days when no open case requires them. These are company-described features rather than an independent assessment of its privacy practices.

In conversation with WION, Aqib Muhammed and Ketan Ramshawala, the founders of the company, said, “AI should not only make it easier to create and share content. It should also help people protect their identity and respond when their images are misused. Technology can support this process, but responsible checks and human judgement must remain important.”

Ultimately, the ability to identify a manipulated image is only one part of the problem. Effective protection also depends on responsible human review, accessible reporting systems and timely action by platforms. As AI-generated content becomes easier to produce and distribute, the challenge is no longer simply how to recognise a fake image. It is how to help people respond when their likeness is misused and how to limit the harm when copies continue to circulate.