oogle might be trying to curb its slop.

Starting Thursday, the Gemini app now tells users whether a photo was created or edited by a Google AI tool when asked the question “Is this AI-generated?” or “Was this created with Google AI?" The tool is currently limited to images, but will soon be extended to video and audio, and will be available in Search at a later date. 

As of now, this identification tool only works against SynthID, Google’s digital watermarking tech that embeds “imperceptible signals” into AI-generated content. After SynthID was introduced in 2023, more than 20 billion AI-generated pieces of content have been watermarked.

Google is also working on verification for Coalition for Content Provenance and Authenticity (C2PA) credentials. This will allow it to detect when content has been generated by other AI tools, such as OpenAI’s Sora or Midjourney.

“Now, as generative media becomes increasingly prevalent and high-fidelity, we are deploying tools to help you more easily determine whether the content you're interacting with was created or edited using AI,” Google said in the announcement.

Generative AI is getting better at creating content that seems realistically human. On Thursday, Google released Nano Banana Pro, its new image generation tool with improved image resolution and text rendering, adding to the growing fray of capable generative models.

And more often than not, people can’t tell the difference between real and fake. And the consequences can be drastic:

• Deepfake audio and video cybercrime has escalated in the past year, with fraud losses reaching more than $200 million in the first quarter of this year alone. 
• AI-generated evidence is increasingly appearing in court, causing judges to question how much they can trust it. 

But Google’s approach to solving this problem should only be one of many, Ben Colman, CEO of Reality Defender, told The Deep View. “This solution, combined with other non-provenance models, creates a ‘Swiss cheese’ approach, where if one method does not stop/catch a deepfake, the other will,” he said.