Launch of the Content Authenticity Initiative White Paper
Today the Content Authenticity Initiative (CAI) spearheaded by Adobe, Twitter and the New York Times publishes its White Paper. WITNESS is a co-author on the paper and has been part of the Working Group developing the standards alongside other groups including TruePic, Microsoft, the BBC, CBC and the Stanford Center for Blockchain Research. The Paper proposes a set of standards for building infrastructure to track the attribution, provenance and manipulation of audiovisual media for creative purposes, journalism and activism. In this blog we assess these “authenticity infrastructure” standards against key criteria for human rights, inclusion, privacy and expression.
WITNESS’s work on authenticity infrastructure
WITNESS brings to this work our experience working with the human rights defenders, civic journalists and critical truth-tellers in communities facing human rights violations. They are often targeted for their work and told their media are lies. They often need to prove what they’ve filmed is true but face significant risks in doing so.
Every day, the stakes increase for how we manage issues of manipulated, fake and deceptive video and audio online, with governments, companies and publics responding to the need to discern truth from falsehood. At the same time, we must be acutely aware of how individual demands to ‘prove it’s true’ and the wider response to perceived mis/disinformation can be weaponized against vulnerable communities and people — and how technical indicators must not become over-intertwined with trust.
A primary focus of WITNESS’s work has been to prepare better and more inclusively for new forms and a growing volume of of media manipulation, misinformation and disinformation. Last year, our pioneering report with Mozilla Fellow Gabi Ivens ‘Ticks or It Didn’t Happen: Confronting Key Dilemmas in Authenticity Infrastructure for Multimedia‘ laid out critical questions for initiatives like the CAI. We know the value of being able to prove whether a video shows a human rights violation or critical documentation of an event. But how will these new proposals protect privacy and ensure vulnerable journalists and activists are not excluded? How will they ensure that infrastructure cannot be weaponized against human rights and vulnerable communities and create further harms and risks?
Authenticity infrastructure issues also relates closely to our work on video as evidence, effective archiving, designing tools for more evidentiary media and ensuring that systemic infrastructure is built with the voices and demands of vulnerable and marginalized communities centered. Much of the early work to think about how to ensure that authenticity tools work for vulnerable communities has come from the non-profit/activist community including groups like WITNESS as well as the Guardian Project, Horizontal, OpenArchive, eyeWitness and others, and leading activists like Harlo Holmes and Gabi Ivens.
In the video below (part of a three part series) my colleague Corin Faife explains more.
What are the key questions about authenticity infrastructure?
We’ve written within our ongoing #TracingTrust series about the critical issues we’d like to see addressed in projects like the Content Authenticity Infrastructure, within ‘verified capture’ tools, as well as in other expanding projects to track media manipulation such as Project Origin
- Whose voices are accidentally or deliberately excluded or chilled? Who needs privacy and anonymity?
- On who is the burden of proof increased?
- What if the ‘ticks’ don’t work or they work too well for media consumers?
- Who abuses the data and the tools?
- What pressures are placed on platforms, journalists and media?
- How do we explain and account for science that is complicated and fallible?
The core question we need to ask now is this: As we move from opt-in and niche authenticity infrastructure to more widespread efforts driven by governments, platforms and public demand… what do we have to do to ensure that authenticated capture or tracking of content authentication and modification at scale doesn’t harm and in fact enhances freedom of expression and trust?
Additionally in the Ticks or It Didn’t Happen report in every chapter we raise specific questions different entities (e.g. developers, platform companies) should ask about each key dilemma we highlight.
These concerns above have been echoed in the ‘expert’ meetings of people with expertise in journalism, fact-checking, human rights, technology and movement leadership that WITNESS has coordinated over the past year in Brazil, South Africa and Southeast Asia to solicit critical feedback on key proposals to fight mis/disinformation.
In this blog we focus on the CAI approach to-date, and highlight areas of promise.
How does the overall CAI approach respond to key questions we’ve raised?
The CAI approach has incorporated many critical elements across goals and focus, underlying principles, focus use cases, and technical architecture decisions.
Among the key technical decisions:
- There is no requirement to tie identity to tracking original media, changes or modifications
- There is the ability to redact media to protect compromising information before releasing it, without this compromising other information
- The system explicitly focuses on providing information on changes, edits and attribution – not on confirming the underlying content is true
- The approach looks to an interoperable open standard, builds on existing technical approaches. It aims to integrate with existing workflows and to not impose undue technical burden to use.
- The system is agnostic to file storage, and does not require data to be stored in the cloud (though some implementations will include this)
Overall, the CAI also emphasizes that “CAI specifications must be reviewed with a critical eye toward potential abuse and misuse of the framework. In addition, CAI specifications must be reviewed for the ability to be abused and cause unintended harm, threats to human rights, or disproportionate risks to vulnerable groups globally” (Principle 2.9).
The white paper also includes specific consideration of workflows specific to human rights activists and civilian witnesses (Workflow 4.3) reflecting these needs around privacy, data protection and redaction.
Here’s how it responds to specific question areas we raise in our #TracingTrust work.
Who needs privacy and anonymity?
Fundamental architectural decisions of the CAI include not requiring persistent identity or linking identity to the original media or changes in the media (5.1.4), allowing people to redact data such as location or blur faces while preserving other attribution data (5.1.6, 7.2), and explicitly stating a principle on protecting privacy (Principle 2.2).
Whose voices are accidentally or deliberately excluded or chilled?
Technology requirements necessary to use a tool, including the device or OS needed, internet connectivity and requirement to use a particular new technology can deliberately or accidentally exclude vulnerable individuals and communities. They may also opt-out because of trade-offs they must manage on privacy, security and visibility around particular media.
The framework explicitly focuses on the impact on global and vulnerable populations, usage by high-public interest individuals like human rights activists and the risks of abusability and misuse (Principle 2.9). It emphasizes the need for interoperable systems (2.4) that “take into consideration the needs of interested users throughout the world” (2.3) and that fit into “existing workflows” (2.5). A ratchet effect is when a new technology creates a negative impression of existing technologies. While avoiding a ratchet effect is almost impossible, the Principles emphasize avoiding ‘unreasonable performance characteristics for implementers’ (2.6). They also foreground avoiding ‘unreasonable technical complexity and cost burden for implementers’ (2.7). These are good steps to ensure that technical requirements do not excessively predetermine who can use these tools. The flexibility noted above to not link identity and to redact some critical data without compromising the whole are critical enabling elements towards ensuring people and communities who must manage visibility and anonymity/privacy can participate.
Governance of the trust list of signing authorities (5.1.3) is as yet undefined. This must ensure inclusion and include global, open-source, and non-commercial projects.
On who is the burden increased to show something is true?
The CAI white paper focuses on avoiding technical barriers to usage. It emphasizes that understanding something has been manipulated or edited doesn’t necessarily tell you it’s true. In our current information environment making the distinction between showing manipulation and asserting authenticity or truth is critical.
As noted in the overarching goals: “CAI specifications should provide a mechanism for the producers and custodians of any given content to assert, in a verifiable manner, any information they wish to disclose about the creation of that content and any actions taken since the asset’s creation. We refer to such information collectively as provenance. CAI specifications should not provide value judgments about whether a given set of provenance data is “good” or “bad,” merely whether the data can be verified as associated with the underlying asset, correctly formed, and free from tampering.”
Providing this type of information can help people contest charges of falsehood, providing an additional source of data. More and more we see claims that genuine media is falsified — the growing ‘it’s a deepfake’ claim that is known as the liar’s divided. CAI information could be a valuable additional signal about potential modifications, not the signal of un-manipulated/manipulated or of the truth of the content.
What if the ‘ticks’ don’t work or they work too well for media consumers?:
As we move forward we need real work on ensuring that the UX built on tools like this does not miscommunicate truth or trust to users. As the CAI white paper notes: “It is extremely important that a consumer of assets not interpret more trust in the presence of valid CAI information than it truly means — specifically that an asset with valid CAI information does not imply anything about the trustworthiness of the content of the asset.” (7.1)
There is good guidance and research emerging in this area, for example from the Partnership on AI’s Manipulated Media project.
Who abuses the data and the tools? What pressures are placed on platforms, journalists and media?
This is an area that we must carefully watch. Infrastructure that tries to protect an interoperable, privacy-protecting framework must not be abused by governments or by platforms. This is particularly true as we see a global trend to crack down on ‘fake news’ for authoritarian purposes and to try surveil people more aggressively. It is also true given the known impact of mechanisms and proposals such as DRM and pre-publication filters.
A tool for helping media and people to better understand and evaluate content they encounter must not become a mechanism for suppressing free expression and media freedom, or an additional tool for data harvesting for corporations. The ability to independently audit CAI-compliant tools to make sure they are not complicit in these approaches is a critical element that is so far missing.
Now what? The rubber hits the road in implementation, building and monitoring
Looking ahead, now the rubber hits the road to uphold these values and standards as the prototypes, tools and tech using the CAI standards are built and implemented. We need to ensure that technology built will enhance free expression and privacy and other human rights values globally. They must enhance the autonomy and agency of vulnerable and marginalized communities, not harm people. Participating companies and developers must continue to integrate these values and questions in the next steps.
One essential next step is to make sure that the input and participation and decision-making from human rights groups and people marginalized from tech decision-making grows rather than shrinks going forward — to ensure that the next steps in this process prioritize equity and rights.