AI Disclosure – A Legal, Ethical & Reputational Issue
It is all fun and games until someone loses an eye.
In some eye-popping media coverage this week, if the three Anthropic researchers who have predicted that AI could drive human extinction within the decade are right, as we play with AI there is more at risk than losing sight.
For every new technology, there’s a tendency to fall in love. We see what is good and often give short shrift to the risks.
As communicators we need to remain balanced. We need to navigate the new currents in the waters we are swimming in.
Although there has been extensive discussion of issues around the adoption of AI, there are emerging concerns that require attention. One of them is disclosure, the focus here.
Legal Disclosures
On August 2, provisions of both the EU AI Act and the California AI Transparency Act came into effect. AI laws are also being adopted in NY, other states, and countries. A commonality across the new legislation is that large AI companies are required to build in disclosures. OpenAI, Gemini, Claude and other LLMs are already doing this.
The EU AI Act goes one step further. It requires brands that publish AI-generated or manipulated content (text, images, audio, or video) to include human- and machine-readable disclosures. As with all legislation, the details matter – see our recent white paper Content Authentication, the EU AI Act, and AI Disclosures.
Tauth Labs is working with law firm Davis+Gilbert on a breakfast roundtable on September 24 supported by CommPRO and H/Advisors for senior industry leaders to discuss the impact of this legislation. We’ll have NY Assemblymember Alex Bores, Adobe’s Head of Content Authenticity, and D+G partners share perspective. In short, if you are using AI to generate text, images, podcasts, or video, you need to know the disclosures required today and those that are coming. Find out more about attending here.
Ethics
There are many new ethical questions and concerns around the use of AI, but I’d argue that for communicators one of the most important is whether or not you disclose its use. I’ve had a number of conversations recently in which this topic has come up in the context of articles and books. Legal disclosure is required under specific circumstances, but for most other situations, the decision to disclose remains personal or based on corporate or agency guidelines.
More disclosure rather than less is in my view generally the best course of action. The more specific the better. When it comes to risk management, documenting how AI was used in a way that is auditable is likely to grow in importance. If in doubt, get legal advice.
Reputational Risk
There are significant new reputational risks associated with the use of AI. The failure to meet disclosure requirements is just one. As some high-profile firms using AI-generated content have found to their cost, the undisclosed use of AI can land you in the headlines and result in serious damage.
Although not the focus here, the broader risk to brands of what we term “shadow content” – malicious content designed to look like it is from you to deceive and defraud – is growing and significant. The disclosure of steps your company is taking to address these risks will become a new corporate best practice.
Concerns about legal disclosures, ethics or reputational risk are not new. But with AI, there are new and specific dimensions to consider.
Importantly, there are solutions. Authenticating digital content, for example, provides ways to address the human- and machine-readable requirements of AI transparency legislation. Building in cryptographic proof of the provenance or origin of content provides audiences with the ability to differentiate between what is authentic and what is fake when integrated into newsrooms, corporate announcements, or research. It gives companies the option of embedding specific AI disclosure information into digital files in a content manifest – that’s the equivalent of a stamp in a digital passport built into content.
We may be approaching the point where the risk of the disclosure of the use of AI is perhaps less consequential than the failure to disclose. In a world in which AI is everywhere, questions around the level of AI use that should require disclosure will be hotly debated for some time to come.
Emerging reputational risks and disclosure requirements around the use of AI go beyond the initial concerns of hallucination or social bias. New issues and concerns will arise. Whether these will be existential for companies and society is a reasonable question.
Amid the fun and games of adoption, a greater awareness of the consequences of AI is emerging. That’s a good thing. We would do well to keep in mind something Warren Buffett said: "It takes 20 years to build a reputation and five minutes to ruin it. If you think about that, you'll do things differently."
