Singapore's Centre for Advanced Technologies in Online Safety (CATOS) will launch a free online tool called Provo in 2026 to help users verify whether images and videos have been altered.

Provo lets publishers, brands, and creators embed provenance labels, which users can view via a free browser extension, with labels resembling nutrition tags that show the original publisher, posting date, and edits before publication.

The Provo initiative is part of CATOS’s broader push funded under SG$50 million through the Research, Innovation and Enterprise 2025 Plan, launched in 2024 to counter online harms like misinformation and AI-generated content.

CATOS director says traditional telltale signs of deepfakes aren’t reliable alone, stressing that content provenance verification must complement deepfake detection technologies.

Officials warn that even high-accuracy systems can fail in large-scale deployment, so careful rollout and ongoing evaluation are essential.

CATOS will publish public education materials to boost AI literacy and advises the public to check source credibility and seek trusted authorities when evaluating content.

Public education efforts will also aim to improve AI literacy, with guidance to verify sources and rely on trusted guidance rather than relying solely on content for judgments.

CATOS operates Sleuth, a deepfake recognition platform that analyzes pixel-level inconsistencies and performs frame-by-frame analysis to detect manipulated faces, audio, or environments; Sleuth is already used by some public agencies and media partners and may be released for public use.

Sleuth's wide public availability is under assessment due to accuracy and infrastructure needs, though it is designed to detect pixel-level anomalies and other manipulated content.

CATOS is also developing Sleuth to analyze pixel-level inconsistencies and conduct frame-by-frame analysis for manipulations across faces, audio, and environments.

CATOS aims to build an ecosystem of trust by engaging content creators and major publishers to adopt Provo and Sleuth, including consideration of Singpass integration to authenticate users.

Trust and user authentication are emphasized for Provo, with potential Singpass verification to prevent impersonation and ensure authentic participation.

The overarching goal is to complement ongoing deepfake detection with provenance-based verification to maintain information integrity, acknowledging that content provenance tech is still in an early stage in the region.

CATOS has signed an MoU with Adobe to explore content provenance technologies and collaborates with partners like CheckMate to enhance fact-checking, hosting the inaugural Online Trust & Safety Forum with over 250 experts.

Provenance tools are viewed as a necessary complement to deepfake detection, described as a cat-and-mouse game given advancing AI-generation techniques.

Labels will provide metadata such as the original publisher, posting date, and edits, and may be removed automatically if content is manipulated by unauthorized parties, while traceability persists when content is reposted or shared.

Provenance markers include the original publisher, first posting date, and edits; automatic removal may occur if content is altered by unauthorized parties.

Adobe partnered with CATOS in 2024 to implement provenance technologies in Singapore, with more Provo details expected in May 2026 and consideration of Singpass integration for secure identity verification.

The Provo rollout follows a 2024 agreement with Adobe to implement provenance tech, with further public details anticipated in May 2026 and possible Singpass-based identity verification.