How do you signal Trust to both humans and AI systems?

This piece first appeared on On Think Tanks. You can read the original article here.

It’s ‘brocante’ season in France – a time when villages and towns open their streets to vendors who set up tables displaying anything from outright junk to pre-loved tools and treasured finds. Mostly, it’s your neighbours keen to shift some of the stuff that’s been gathering dust in attics and garages, but there are often professional ‘brocanteurs’ too, including restorers and antique dealers. Haggling is the modus operandi, buyer beware, the watchword.

You know which neighbour to trust, but what about that brocanteuse with a persuasive patter who’s offering some attractive-looking silver fish-knives and a gold necklace which she says have been in her family for generations. How do you know if they’re real or fake?

Well, help is at hand because the genuine article will carry a mark that confers trust. A horse’s head stamped on the gold necklace will not only tell you it’s made of 18ca gold, but it was made between 1838 and 1919. A boar’s head on those elegant fish knives will tell you that they are indeed made of silver, but were imported.

These marks don’t say anything about an object’s beauty or utility; that’s not their job. Their role is to tell a trust story about the object, verifying the authenticity of the seller’s claim about the quality of the metal used and the object’s provenance – the what, the when, and often the who too. Hallmarks – one of the earliest forms of trust infrastructure, long before logos and brands – work because they are standardised, indelible and travel with the object wherever it goes. They persist, conferring trust for as long as the object exists, including long after the original maker and vendor have disappeared.

Now you’re back at your desk. This time, the treasure you’re looking for is information. Like the French brocante, there are hundreds of vendors setting up their stalls online. Some are like your neighbours, you know and trust them, but many you’ve never heard of. How can you know whether you can trust them? Where’s the ‘mark’ that will tell a story about their trustworthiness?

Then there’s that email which arrived yesterday with a PDF attachment, which the sender says is a copy of that IGO report which was all over the news yesterday. In a world where fake reports circulate, how do you know if it’s genuine? And if it is a copy of the original, how can you be sure that no one has altered the data, conclusions or recommendations before it reached you?

Crucially, how will AI systems know which content to trust and which to discard when crafting responses to user prompts?

What’s needed is a hallmark for today’s precious objects, digital information. Which is why we’re launching the Trustmarc Initiative.+

Just as hallmarks tell users that a gold necklace can be trusted, so Trustmarcs will tell users that a published item can be trusted.

A hallmark is backed by a country’s independent assay office. They set the standards and approve a network of trusted entities who can stamp the objects. Similarly, a Trustmarc will be backed by an independent Trustmarcs Standards Council – which is being formed now. They will set the standards and approve a network of trusted entities who will badge digital publications. Your organisation can be one of them.

Each Trustmarc badge is backed by a digital Trustmarc Certificate stored in a secure online registry. Click the badge to verify the document’s authenticity and provenance — who created it, where it came from, and what’s been done to it.

For think tanks, trust in your publications is vital. Readers may disagree with your conclusions and recommendations, but they need confidence that a publication genuinely comes from your organisation, that it has not been altered, and that the evidence and review processes behind it are transparent. That’s why every Trustmarced item will have a tamper-evident layer (provided by C2PA’s Content Credentials standard), so they can be confident that what they’re reading is exactly what was published.

Trustmarc is not intended to replace peer review, editorial review or organisational reputation. It’s designed to work alongside your brand and reputation, delivering an additional layer of verifiable information about authenticity, provenance and integrity for humans – and machines.

The Trustmarc badges and certificates will be added to an item’s metadata so search engines and AI systems can also know what to trust and why. The C2PA standard is already backed by major players, including Microsoft, Google, OpenAI and the BBC, and others are expected to follow.

The Trustmarc Initiative is starting to be rolled out now – the underlying technology is being built by the same team who build Policy Commons. We’re now looking for up to a dozen development partners – drawn from across the information landscape – to help us test workflows and be among the first to deploy Trustmarcs on their content. If you’d like to learn more about being a development partner, please get in touch with Toby Green by the end of July.