TIME names Google’s SynthID a Best Invention of 2026, highlighting AI watermarking and verification tools

TIME has named Google DeepMind’s SynthID among its Best Inventions of 2026, elevating a technology aimed at addressing a growing trust problem on the internet. As generative AI systems make images, audio, video, and text look increasingly real, the ability to determine how content was produced is becoming central to online navigation and public confidence.

In a Google blog post announcing the recognition, the company said SynthID works by embedding an invisible, durable digital watermark into AI-generated content at the moment of creation. The watermark is designed not to alter the quality of the media, according to the post, while still allowing the origin of content to be identified later.

Google provided scale figures to show how widely the approach has been exercised. To date, it said SynthID has watermarked more than 180 billion images and videos and 240,000 years of audio. Those totals are presented as accumulated tracking of content that has gone through the watermarking step, rather than as an estimate of how much AI-generated media on the wider web carries a comparable marker.

Beyond the watermark itself, the post tied SynthID to a broader set of verification tools. Google said it has built detection capabilities into Google Search, the Gemini app, and Chrome to help people identify AI content while they browse. In addition, Google said the SynthID Detector can be used to check content from Google and from partners.

TIME’s framing, as described by Google, links the technology to “knowing how content was made,” positioning watermarking and verification as a practical layer for web users. The company’s account suggests the goal is not only to authenticate individual files but also to provide transparency and context in the interfaces where people already encounter content.

Google also used the announcement to connect SynthID to another emerging use case. The company said it recently announced SynthID Bio, described as bringing SynthID’s watermarking approach to synthetic biology. The post did not provide further technical detail on how watermarking would be represented or detected in biological data, but it indicates Google is looking to apply provenance methods beyond traditional media pipelines.

For Alphabet, the decision to emphasize SynthID fits into a larger pattern in the generative AI market: moving trust and verification upstream into model output and downstream into consumer products. Google’s approach, as laid out in the post, combines an embedded marker with detection tooling inside major distribution channels, including a search engine, a messaging and assistant app, and the Chrome browser.

That integration matters because it changes how verification is experienced. Rather than making users hunt for third-party tools, the company says the checking is built into the products where content is discovered. In practical terms, the announcement implies that the value of watermarking depends not just on embedding indicates, but on making detection accessible at the moment of viewing.

Still, the post and TIME recognition leave several questions unanswered. It does not disclose how well the watermark performs under heavy editing, compression, or re-encoding across all media types. It also does not quantify adoption by creators, partners, or platforms, nor does it provide details on whether detection is available for all content sources or only for content generated through systems that apply SynthID.

Looking ahead, the next test for SynthID will likely be real-world coverage as AI content spreads and as verification tools encounter a wider mix of sources and workflows. Observers will also watch how SynthID Bio develops, including what kinds of synthetic biology outputs are targeted for provenance, and whether similar “embedded at creation” principles can be translated from media to biomedical data.

4 to 7 concise facts grounded in The announcement and source list. Plain text only: no markdown links, no URLs, no parenthetical citations.

3 to 5 market/business implications, phrased cautiously if the source is thin. Plain text only: no markdown links, no URLs, no source citations.

sourceTrail must include the original source URL and every researched URL that materially support the story.