Since iOS 27.2, some of the AI tools in Apple's Photos app leave a mark nobody can see. A support note on Apple's Malaysian site, published on 5 November, says images created or edited with certain Apple Intelligence tools now carry an invisible SynthID watermark that a detector can read. It arrives in a country where deepfake complaints to MCMC reached 7,967 between January and 15 June this year.
Quick facts
- Tools that add the watermark: Clean Up, Extend and Spatial Reframing in Photos, plus Image Playground and Image Wand.
- Software: iOS 27.2, iPadOS 27.2, macOS 27.2 and visionOS 27.2. Apple lists iOS 27.2 as released on 26 October.
- How to check an image: upload it at synthid.com.
- Deepfake complaints to MCMC: 7,967 from 1 January to 15 June 2026, against 3,612 in all of 2025.
What Apple changed
Three of the five tools are editing tools inside Photos. Clean Up removes objects from a shot, Extend widens a photo past its original edges, and Spatial Reframing moves the viewpoint after the picture is taken, as described on Apple's Malaysian Apple Intelligence page. The other two, Image Playground and Image Wand, make new images.
Apple says the watermark is imperceptible and does not change how the image looks. Photos edited before the update carry no watermark unless they are edited again with one of the five tools. So a real holiday photo with a stranger cleaned out of the background now carries the same kind of mark as a picture made from scratch.
Apple lists iOS 27.2 as released on 26 October on its Malaysian security updates page, and we covered that update at the time. Apple Intelligence needs the device and Siri language set to one of the languages Apple lists, English among them. The cheapest iPhone in the device list on that Malaysian page is the iPhone 17e, from RM3,299 on Apple's Malaysian store.
How to check a photo
Apple points to synthid.com: open it in a browser, upload the image and follow the steps to see the result. SynthID is a watermarking system made by Google DeepMind.
Where Malaysia's rules come in
MCMC's Risk Mitigation Code took effect on 1 June under the Online Safety Act 2025. In its statement announcing the code, carried by Bernama on 22 May, MCMC listed labelling of manipulated content among the measures platforms must take, and called the approach outcomes-based, leaving platforms room to choose how they meet it. Platforms that fail their obligations under the code can be prosecuted and, if convicted, face a fine of up to RM1 million along with additional financial penalties of up to RM10 million, Deputy Communications Minister Teo Nie Ching told the Dewan Rakyat on 30 June, as reported by Bernama. We covered the same law's under-16 rule earlier this week.
In the same answer she gave the complaint numbers: 917 in 2024 and 3,612 in 2025, the two full years, then 7,967 from 1 January to 15 June 2026. The periods differ, so this is not a like-for-like rate, but five and a half months of 2026 already brought more than twice the complaints of the whole of 2025. More than 11,600 items of false or deepfake content have been taken down since 2024 after MCMC asked platforms to remove them.
Apple's watermark and the code are separate things. Apple's note does not mention Malaysia's rules, and we found no statement from MCMC or any platform on whether it reads SynthID watermarks when deciding what to label. A hidden mark only helps a platform label an image if the platform checks for it.
To see whether a photo already carries the mark, upload it at synthid.com. Apple says images edited before the update do not get its watermark until they are edited again with one of the five tools it lists.
