HarmBlock
Yes. No AI system is 100% accurate all of the time.
There are two main kinds of errors:
HarmBlock is pioneering a new approach to on-device child safeguarding, designed to provide broader protection across a child’s digital environment. Recognising visual content in everyday situations is complex. Images may be blurred, cropped, compressed, partly hidden or affected by lighting, movement, screen transitions, glare and other distortions.
We are continually refining the technology and welcome reports of any issues that can help us reduce errors and strengthen protection over time. If you think HarmBlock has made a mistake, please report it using our vulnerability disclosure form.
HarmBlock is designed to block sexual content that is clearly not suitable for children, with a particular focus on nudity and explicit sexual activity.
However, the AI is designed to err on the side of caution, so there are some borderline cases where it may still step in.
Examples can include people in underwear, lingerie or swimwear with a high degree of skin exposure, provocative imagery that looks visually similar to explicit content, or frames where clothing, lighting, blur or angle make it hard to tell what is happening.
This is part of the trade-off we’ve chosen: we would rather be slightly over-cautious in some situations than risk allowing clearly inappropriate sexual content through for children.
HarmBlock is designed to minimise the chance of harmful sexual content getting through, but no system can be perfect in every situation.
If your child tells you something should have been blocked, the most helpful next step is to let us know what happened so we can investigate patterns and improve future performance. You can do this by reporting it here. (link to Vulnerability disclosure form)
This kind of feedback is exactly what helps us strengthen detection and reduce repeat issues over time.
This technology will continue to evolve as we improve it and test it against new risks to further reduce the risks of being online with every new update. HarmBlock was created with a clear purpose: to reduce harm and help protect children.
It is designed to disrupt harmful situations, intervene at critical moments and reduce the likelihood of a child being exposed to, creating or sharing harmful content in an otherwise unprotected digital environment.
We must reduce children’s exposure to harm. Our approach is to deploy HarmBlock responsibly, measure its performance honestly and improve it continuously, rather than wait for a standard of perfection that no safety technology can meet.