ETSI Issues Standard to Curb Harmful Deepfakes

Technical Specification TS 104 275 sets out security and governance measures for generative AI systems to reduce deepfake misuse.

Updated on Nov. 11, 2026 in Artificial Intelligence

ETSI Issues Standard to Curb Harmful Deepfakes

Live Poll

Do you trust AI companies to prevent their tools from creating harmful fake images?

ETSI has released Technical Specification TS 104 275, aimed at reducing the risk that generative AI systems produce harmful deepfakes. The specification was developed by ETSI's Technical Committee on Securing Artificial Intelligence and builds on baseline cybersecurity requirements in EN 304 223.

Why it matters

Harmful deepfake material can be created by misusing legal, commercially available software, making prevention a practical security concern. The specification supports objectives of the EU AI Act while extending security expectations beyond systems classified as high-risk.

TS 104 275 specifies five measures and includes test scenarios with pass/fail criteria layered on the EN 304 223 baseline. Whether developers will adopt these tests at scale, and what effect they will have on deepfake prevalence, is not yet established.

The players

ETSI

The European Telecommunications Standards Institute is a European standards organization that develops technical specifications for information and communications technologies.

Alex Cadzow

Cadzow serves as Rapporteur and Secretary of ETSI's Technical Committee on Securing Artificial Intelligence, which drafted the specification.

The details

The specification treats deepfake prevention as a security and governance challenge across the full AI system lifecycle. Recommended safeguards include content filtering, labelling of AI-generated content and user warnings where appropriate.

Timeline

  1. November 11, 2026: ETSI publishes the specification and related coverage appears.

Big Picture

The specification extends the security objectives of the EU AI Act beyond the systems the Act classifies as high-risk, shifting deepfake prevention from after-the-fact detection toward built-in safeguards during development and deployment.

The safeguards described, including content filtering, AI-content labelling and user warnings, could change how generative tools behave for everyday users. Developers adopting the specification may add these protections to products used for creating images, audio and video.

The takeaway

Deepfake risk is increasingly being addressed through built-in design and governance rather than only detection after content is made. Anyone building or using generative tools should watch how these safeguards are adopted.

Further reading

Explore more on Artificial Intelligence developments and governance standards.

Source note: This article includes information reported by SecurityWorldMarket.

Live Poll

Do you trust AI companies to prevent their tools from creating harmful fake images?