The proliferation of images and videos generated through artificial intelligence makes it increasingly difficult to distinguish between authentic content and falsifications.

Media Integrity and Authentication (MIA)

Media integrity and authentication (MIA) techniques aim to verify the source and history of digital content. The Coalition for Content Provenance and Authenticity (C2PA) leads the development of standards in this field, alongside complementary methods such as watermarks and fingerprinting.

Microsoft Research has identified four key factors:

  • Increasing spread of synthetic media.
  • Legislation on verifiable provenance.
  • Pressure for clear authentication signals.
  • Awareness of attacks on the vulnerabilities of authentication systems.

The Media Integrity & Authentication: Status, Directions & Futures report analyzes the real limits, edge cases, and emerging attack surfaces for MIA methods.

Key Findings

The research defined two new concepts:

  • High-Confidence Provenance Authentication: accurate verification of the origin and modifications of an asset.
  • Sociotechnical Provenance Attacks: attacks aimed at inverting authenticity signals.

The report identifies four promising directions for strengthening media authentication, including:

  • Providing high-confidence provenance authentication.
  • Mitigating confusion from sociotechnical attacks.
  • Enabling more trusted provenance on edge devices.
  • Investing in research and policy development.

Implementation and display choices affect the interpretation of provenance signals. The use of a C2PA manifest signed in a secure environment enables high-confidence validation, further strengthened by linking an imperceptible watermark to the manifest. Fingerprinting supports manual forensics but does not enable high-confidence validation.

MIA methods are vulnerable to sociotechnical attacks that can deceive the public. Overlaying secure provenance and imperceptible watermarking can mitigate the impact of such attacks. UX design that allows users to explore manifest details can support fact-checking activities.

Implementing a secure enclave in the hardware of offline devices is essential to make the provenance of captured images, audio, and video more reliable. UX and display are promising areas for future research, with in-stream tools that show provenance information and distinguish between high and low confidence signals. Stakeholders should conduct ongoing analysis to identify and mitigate weaknesses through technical approaches, policies, and laws.