AI is becoming one of the most useful tools ordinary people can access. It can help someone learn a language, explain a difficult medical paper, make software more accessible, support coding and design, and give a small team capabilities that once required an entire department.
That is the optimistic side of AI. It deserves attention.
The uncomfortable side deserves just as much. The same progress that makes image generation more convincing also makes it easier to create fake, sexualised, or humiliating images of real people who never agreed to be part of the process. A public picture can become source material for harassment, impersonation, or a non-consensual intimate image.
This is where the conversation needs more balance. Using a flower, sticker, or another design on part of the face can be a useful precaution when someone wants to share a picture online. It may make casual copying or recognition more difficult. But it is not the complete solution. Modern AI can still manipulate, reconstruct, or combine visual details in ways that do not depend on one part of a picture. The deeper problem is that AI systems and online platforms are still too easy to misuse.
The good and the bad are arriving together
AI is not automatically good or bad. It is a capability shaped by the people who build it, the data used to train it, the product choices around it, and the way people use the final system.
The benefits are real:
- Faster research and summarisation
- Translation and accessibility support
- Tutoring and personalised learning
- Creative tools for people who cannot afford specialist software
- Assistance with coding, design, writing, and repetitive work
- New ways to explore ideas before investing significant time or money
The risks are real too:
- Non-consensual intimate imagery and sexual deepfakes
- Impersonation and identity fraud
- Harassment and blackmail
- Misinformation that looks like a picture or video
- Biased decisions caused by incomplete or unrepresentative data
- Privacy loss when personal information is uploaded without understanding how it is stored or used
The difficult truth is that better image quality improves both legitimate creativity and malicious deception. We cannot keep only the useful half by telling individuals to be less visible.
LinkedIn creates a genuine privacy trade-off
LinkedIn deserves a specific mention because a professional profile picture can be useful. In many industries, a real headshot helps colleagues recognise someone, makes networking easier, and gives a profile a basic sense of identity. For some people, not having a picture can even create an unfair professional disadvantage.
At the same time, a profile image visible to other people can potentially be copied, screenshotted, or reused outside the context in which it was shared. That is not a reason to blame the person in the picture. It is a reason to treat public professional images as public-facing data and to improve the controls around them.
Review LinkedIn’s current visibility options rather than assuming a profile image is private. A practical approach is to use an image that is good enough for recognition but not a high-resolution original, keep the original file private, and avoid publishing extra personal details that make impersonation easier.
That is risk management, not a moral requirement. No woman should be told that she must cover her face to deserve safety online. Women and girls are often targeted disproportionately by sexualised abuse, but they should not be asked to carry the whole burden of preventing it.
A small precaution is useful, but AI must be challenged
A flower, sticker, blur, or decorative mark on part of the face can be a sensible precaution when sharing a picture online. It may reduce casual recognition and make simple copying less useful. People should be free to use that kind of protection if it makes them more comfortable.
But the precaution cannot carry the full burden. AI systems can reconstruct, edit, or combine visual information, and online platforms can allow abusive material to spread quickly. The stronger question is not only, “What can a person do to protect a picture?” It is also, “Why are AI tools and platforms still making it so easy to create, distribute, and profit from abuse?”
A responsible system can combine:
- Clear rules against sexualising real people without consent
- Checks for prompts that target identifiable people
- Restrictions on face or identity manipulation
- Output moderation that blocks or quarantines sexual content
- Provenance signals that help people understand how media was made
- Abuse reporting, human review, rate limits, and meaningful enforcement
- Testing with real abuse scenarios rather than relying on a single benchmark
No individual filter is perfect. Safety has to be layered.
Tools that can reduce the risk
Developers building image products do not have to start from nothing. Several tools can help, although none can guarantee that an explicit image of a real person will never be generated.
- The open
CompVis/stable-diffusion-safety-checkermodel on Hugging Face is designed to identify NSFW images. It is useful as an output screening component, but it is not an identity-consent system. - Hugging Face Diffusers documents safety-oriented pipelines such as Safe Stable Diffusion. These can miss content, over-block benign content, or fail when removed or bypassed.
- Google Cloud Vision SafeSearch and Amazon Rekognition’s
DetectModerationLabelsprovide moderation paths for images. They help with detection and review queues, not proof of consent. - C2PA Content Credentials can record information about an asset’s origin and editing history. Provenance does not prevent abuse by itself, but it can make synthetic or edited media easier to investigate.
- StopNCII.org can create a digital fingerprint of an intimate image on the person’s device and share the hash with participating platforms. It is a response tool, not a substitute for prevention or legal advice.
The important distinction is this: NSFW detection asks, “Does this output look sexually explicit?” Consent protection asks, “Is this person identifiable, and did they agree to this use?” A serious product needs both questions, plus a way to respond when the system gets them wrong.
What ordinary users can do without disappearing
- Review who can see a profile and whether it is discoverable through search
- Share a suitably sized copy of a professional picture instead of the highest-resolution original
- Keep personal contact details, addresses, travel plans, and identity documents away from public posts
- Use two-factor authentication and unique passwords
- Keep a record of impersonation or abuse: URLs, timestamps, messages, and screenshots where safe and lawful
- Report the material through the platform’s abuse process and do not re-share it
- Tell a trusted person. Support helps victims make decisions
These steps are sensible for everyone. They are not instructions for women to become less visible so that other people can behave badly with fewer consequences.
The moral boundary
AI should make people more capable, not make them easier to violate.
Model builders must design stronger safeguards. Platforms must make reporting and removal effective. Employers and communities must support targets instead of questioning why they posted a picture. Lawmakers need rules that recognise synthetic abuse. Ordinary users must refuse to create, request, save, or circulate sexualised images of real people without consent.
The right moral is not “girls should cover their faces.” If a person’s image is used to create sexual or humiliating content without permission, the wrongdoing belongs to the person who made or spread it—and to the systems that knowingly allow it to continue.
Public visibility should never be treated as consent.
Author note: I am writing this as an AI enthusiast who believes the technology’s benefits are worth developing responsibly.
Top comments (0)