Resemble AI launches deepfake detector & threat report

Resemble AI has launched a deepfake threat report and two free detection tools intended to help users verify digital media in real time.

The new products include a Google Chrome extension for scanning image, video and audio content, and an X bot that lets users check suspicious posts without leaving the platform. Resemble AI also introduced three features for corporate users: multimodal watermarking, a zero-retention mode for submitted media, and a reverse image search tool designed to identify synthetic content without a prior digital footprint.

The release comes as concern grows over AI-generated media spreading across social platforms, news feeds and digital communications. Resemble AI cited Europol estimates that as much as 90% of online content could contain some form of AI-generated material by the end of 2026.

Threat report

The report is based on what Resemble AI described as a proprietary database of verified deepfake incidents compiled from global media coverage. According to the company, the study recorded 1,567 unique incidents in 2025 from 3,253 news stories, with each case classified by attack type and target category.

Among the findings, non-consensual intimate imagery and child sexual abuse material accounted for 20% of verified incidents. Resemble AI said nearly USD $1.3 billion in confirmed fraud losses were linked to generative AI deepfakes, while about 80% of incidents disclosed no damage figure.

The report also found that the average corporate deepfake incident remained in the news cycle for 3.5 years, suggesting reputational effects can persist long after the original episode has faded from view.

Free tools

The Chrome extension lets users scan media on websites with a single click. It displays results through a colour-coded badge system: green for authentic content, red for AI-generated content and yellow for uncertain results.

The extension can also provide frame-by-frame analysis for video and segment-by-segment scoring for audio. It is designed to work across a range of websites and social platforms, including X, Reddit, Instagram, TikTok, Facebook, LinkedIn, Vimeo and Twitch.

The X bot works within posts on the platform. Users can tag the @resemble_detect account alongside the phrase "is this fake?" to request an automated image or video scan, with the result returned in the thread.

Resemble AI said the bot is aimed at journalists, researchers and members of the public who want to assess potentially misleading content without leaving the platform.

Corporate features

For businesses, Resemble AI introduced multimodal watermarking to sign content at the point of creation across audio, image and video. The system is intended to provide a chain of custody by embedding invisible signatures into files as they are generated.

It also announced a zero-retention mode aimed at customers in sectors such as finance and healthcare that face legal or compliance concerns about storing sensitive media in the cloud. Under this model, submitted files are analysed and then immediately purged, according to the company.

The third feature, reverse image search, is intended to help detect what the company called "zero-day" synthetic media. Resemble AI said it searches the web for matching images, checks known debunked content and traces source material to help identify fakes that may not fit established statistical patterns.

Zohaib Ahmed, Chief Executive Officer and Co-Founder of Resemble AI, said the spread of synthetic media has widened the challenge beyond large organisations and technical specialists.

"For years, the industry focused on making AI-generated voice, image and video more realistic," Ahmed said. "We started by building voice AI models, so we understand how these systems work and how they can be weaponised. Multimodal generative AI security is now foundational for enterprises, employees and everyday people trying to navigate a world where more content is now synthetic."

Founded in 2019, Resemble AI develops models for generating and detecting synthetic media across audio, video and image formats. The company said its open-source text-to-speech model has surpassed 5 million downloads on Hugging Face, and that its detection model has been trained on data from more than 160 AI models.