Google Opens AI Content Detector Worldwide as Deepfake Scams Grow

Google has expanded access to its SynthID Detector, allowing people worldwide to check images, videos and audio files for signs of AI-generated content. The tool, previously available in an early version to media professionals, is now accessible globally in English as technology companies face mounting pressure to help users distinguish authentic media from synthetic content.

Pushmeet Kohli, vice president of Science and Strategic Initiatives at Google DeepMind, announced the expansion in a blog post, saying the tool was designed to give people more context about the digital material they encounter. The rollout comes as AI-generated videos, images and voices become increasingly realistic, creating new opportunities for creative work but also making impersonation, misinformation and online fraud harder to detect.

SynthID works by embedding invisible digital watermarks into supported AI-generated images, videos and audio. Unlike a visible label, the watermark is designed to remain imperceptible to viewers and listeners. The detector scans uploaded files for these identifying signals and can indicate whether a watermark is present, helping users establish whether content carries markers associated with participating AI systems.

Google says the detector supports SynthID-watermarked content created using its own AI tools and those of partner companies, including OpenAI, NVIDIA and Kakao. Apple is expected to join the initiative, according to Kohli. The wider support could make the tool useful beyond Google's own products, although it remains dependent on the presence of a compatible watermark rather than attempting to identify every possible form of AI-generated media.

The scale of adoption is already substantial, according to Google. Kohli said the company had watermarked more than 180 billion images and videos, alongside the equivalent of 240,000 years of audio content. Google introduced SynthID in 2023 and has since incorporated verification features into services including Google Search, the Gemini app and Chrome. The company says those services collectively handle more than one million requests a day. These figures are company-reported measures of use and do not independently establish how accurately every file can be verified.

The detector is not, however, a universal AI-content checker. It searches for SynthID signals in supported media, meaning it may not identify content generated by tools that do not use the technology. A file without a detectable watermark should not automatically be considered authentic, nor does a positive result by itself establish that a video or image is deceptive. AI-generated material can be legitimate, while genuine media can also be edited or presented misleadingly.

That distinction is especially important in Nigeria, where fabricated videos featuring public figures, including ARISE News anchors, have been used in financial scams. Such material can falsely associate familiar faces and trusted media personalities with investment offers or other schemes designed to persuade people to hand over money or personal information. The presence of a recognizable presenter in a video is not proof that the person endorsed the message, and viewers may need to check the original source or contact the organization involved before acting on a claim.

The wider challenge extends beyond individual scams. Synthetic media can be used to impersonate business leaders, spread false statements attributed to public officials, manipulate political debate or create fraudulent recordings of family members and colleagues. As voice and video generation tools become more accessible, criminals may be able to produce convincing material at relatively low cost, while the people targeted may have little time or technical expertise to establish whether it is genuine.

For journalists and fact-checkers, SynthID offers another piece of evidence to use when examining suspicious content. But reliable verification still requires a broader process: tracing where a file first appeared, checking whether the person or organization involved has confirmed it, comparing versions of the material and consulting credible sources. A watermark check can support that work, but it cannot replace it.

Google’s decision to make the detector publicly accessible reflects a wider effort across the technology industry to establish clearer signals of digital content’s origins. Other approaches, including content credentials that record aspects of a file’s creation and editing history, aim to help users understand how media has been produced or modified. No single system can solve the problem alone; the usefulness of these tools depends on adoption, interoperability and public understanding of their limitations.

The expansion gives ordinary internet users another way to question what appears on their screens before believing or sharing it. Its real value will depend not only on how many files carry SynthID watermarks, but also on whether people understand what the results mean. As synthetic media becomes more convincing, protecting digital trust will require a combination of technical safeguards, responsible platform practices, public awareness and the habit of verifying extraordinary claims before acting on them.