Resemble AI adds multimodal watermarking to secure AI content

Resemble AI expanded Resemble Watermarker from audio-only to image, video and text on June 24, 2026, giving enterprises a way to embed tamper-resistant provenance marks across more AI-generated content. The launch comes as regulators tighten disclosure rules and organizations face rising deepfake, impersonation and fraud risk.

Why it matters: - Resemble AI is trying to give organizations a verifiable chain of custody for AI-generated content across audio, video, image and text. - The move targets a growing gap between the volume of synthetic media and the limited ability of many enterprises to prove what content is authentic. - The company is positioning watermarking as a complement to deepfake detection, not a replacement for it.

What happened: - Resemble AI expanded Resemble Watermarker from audio-only watermarking to full multimodal coverage on June 24, 2026. - The update adds image, video and text support through PerTh Multimodal, a rebuild of the model behind the company's audio watermarker. - Resemble AI says the new version embeds imperceptible, tamper-resistant marks across every content type an enterprise generates. - Resemble AI also says the audio watermarker remains available as open source.

The details: - Resemble Watermarker embeds the mark directly into the content, rather than relying only on metadata that can be stripped during re-uploading, compression or re-encoding. - A watermarked asset can identify who issued the content, whether the content is AI-generated, AI-altered or authentic, whether it has been tampered with since creation, and whether other provenance marks are present. - Resemble Watermarker reads SynthID and C2PA markers. - Resemble Watermarker writes and signs C2PA credentials. - The system can also add an explicit attribution mark that encodes issuer identity, content date and authenticity status. - Deployment options include on-device, VPC and air-gapped environments. - Resemble Watermarker is available through a self-serve API at resemble.ai. - Enterprise licensing, including VPC and on-premises deployment, is available through Resemble AI's enterprise sales team. - Resemble Watermarker integrates with Resemble Detect for deepfake detection.

Between the lines: - Resemble AI is betting that provenance will become as important as detection as synthetic media becomes easier to create and harder to trace. - The company cites a June 2026 Gartner report naming deepfakes one of four critical threats where enterprise defenses are most overmatched. - Resemble AI also cites The 2025 Resemble AI Deepfake Threat Report, which found brand and reputation attacks generated 156.3 billion impressions last year and synthetic media drove 296.4 billion impressions across all categories. - Research presented at ACM in 2025 found that only about 38% of AI content generators watermark their outputs adequately. - Zohaib Ahmed, CEO and co-founder of Resemble AI, said watermarking answers who made something and whether it changed, while deepfake detection answers whether content is AI-generated or altered.

What's next: - Resemble AI is aiming at the compliance wave tied to AI content labeling rules in the EU and other jurisdictions. - The company says its system maps to the EU AI Act's Article 50 transparency obligations, which take effect August 2, 2026. - The company also points to China's mandatory labeling rules, which took effect September 1, 2025. - Resemble AI expects more jurisdictions to adopt similar requirements as machine-readable provenance becomes the default approach.

The bottom line: - Resemble AI is turning watermarking into a broader AI trust layer, with one product now designed to cover provenance, authenticity and regulatory labeling across the main forms of synthetic media.