Content Credentials guidance

Content Credentials: Strengthening Multimedia
Integrity in the Generative AI Era

Executive summary

With the rise of advanced tools that enable the rapid creation, alteration, and distribution
of images, videos, and other digital content, there are many ways to manipulate what
people see and believe. The ability to manipulate media is not new, but the accessibility,
speed, and quality of these modifications today, powered by artificial intelligence (AI)
and machine learning tools, have reached unprecedented levels and may not be caught
by traditional verification methods. While many people are using generative AI to create
(or modify) and distribute useful creative content faster, it is important to acknowledge
the risks and potential harms of the technology when used for malign purposes. For
example, malicious actors can use manipulated or fully synthetic media in cyber threat,
criminal, or other malign activity against organizations and individuals to impersonate
and misinform. In addition, there are broader societal risks around loss of trust online
that can impact not just individuals and businesses, but whole communities. Due to this
widespread ability to convincingly create or modify media, verifiable media is becoming
critical for ensuring transparency by providing context about the media’s provenance
and integrity with an effective, secure, and robust technical standard.

This cybersecurity information sheet, authored by the National Security Agency (NSA),
Australian Signals Directorate’s Australian Cyber Security Centre (ASD’s ACSC),

* * *

TLP:CLEAR

Canadian Centre for Cyber Security (CCCS), and United Kingdom National Cyber
Security Centre (NCSC-UK), discusses how Content Credentials (especially Durable
ones) can provide transparency for the provenance of media, raises awareness of the
state of this solution, introduces recommended practices to preserve provenance
information, and emphasizes the importance of widespread adoption across the
information ecosystem.

Introduction

The widespread availability of AI and machine learning tools, including generative
models and deepfake technologies, makes it possible for anyone to convincingly create
and/or modify media with minimal effort, low cost, and increased realism. This rapid
evolution poses a significant challenge for traditional verification methods, which may
struggle to keep up with the growing sophistication and scale of these technologies. As
a result, the accuracy and effectiveness of verification methods are increasingly under
strain, leaving consumers more vulnerable to misinformation and influence operations.
The abuse of AI-generated media1 also represents a significant cyber threat to
organizations, including through impersonation of corporate officers and the use of
fraudulent communications to enable access to an organization’s networks,
communications, and sensitive information. Some of these threats were described in the
previous joint cybersecurity information sheet

... (truncated, 63 KB total)