Transparency Is the New Competitive Moat
2026-07-10 · 5 min read · Janaina Maia
Three stories this week. Same thread. Different camps.
Google announced it will now label which ads on its platform were generated or substantially edited by AI. Not bury the label in a policy page. Put it on the ad itself, where the person seeing it can actually find it.
The New York Times told a court that OpenAI deliberately hid evidence in their copyright trial. Billions of logs that should have been preserved were deleted. OpenAI says it was a storage management decision. The court will decide whether to believe that. But the pattern is clear: when transparency threatens your business model, your instinct is to obscure.
And TechCrunch reported that Anthropic's new Claude feature is quietly selling users on AI. The feature, which helps people accomplish tasks with AI assistance, was designed to expand adoption. That is fine on its own. But the word "quietly" is doing a lot of work in that headline. When a company that brands itself on transparency and safety launches a feature designed to increase AI dependency without making that purpose explicit, you have to ask: is this building trust, or is it building dependence and hoping nobody notices the difference?
Three companies. Three approaches to the same question: how much do you let the user see?
The split is real
Right now, the AI industry is sorting itself into two camps. One camp treats transparency as a product feature. The other treats it as a regulatory obligation to be minimised.
Google's ad disclosure is in the first camp. It is not perfect. The label is small. It applies to ads that are "substantially" AI-generated, which leaves room for interpretation. But it is a start. It says: we will tell you when a machine made the thing you are looking at, and we will put that information where you can see it.
OpenAI's evidence handling is in the second camp. The company that built its brand on making AI safe and beneficial deleted logs that were central to a copyright case brought by the people whose work it trained on. Whether or not you believe it was intentional, the effect is the same: the people who should have been able to see what happened cannot see it.
Anthropic's quiet feature sits somewhere uncomfortable in between. The feature itself is useful. Making AI more accessible to people who have not tried it yet is a legitimate product goal. But Anthropic's entire brand is built on being the responsible, transparent alternative. When that same company launches a feature designed to expand AI usage and does not make the expansion goal explicit, it is not building trust through transparency. It is building usage through convenience and hoping the user does not notice the difference.
Why this matters for product design
Most product conversations about AI transparency focus on compliance. GDPR labels. Terms of service updates. AI Acts and executive orders. These matter. But the competitive moat is not compliance. The competitive moat is trust.
Trust in AI products is built in three layers.
Layer one: Can I see what the AI did? This is the basics. Can I tell whether an ad, an article, an image, or a recommendation was generated by a machine? Right now, most AI products do not tell you. Google is starting to. That is a product decision, not a legal requirement.
Layer two: Can I see what the AI is doing right now? This is the active transparency layer. When Claude is helping me write an email, is it also quietly trying to make me more dependent on AI? When an agent is accessing my files, can I see what it is reading and why? When a model is trained on my data, can I find out? The difference between "we are helping you" and "we are hooking you" is often just the transparency of the intent.
Layer three: Can I control what happens next? This is the agency layer. Not just "I can see it" but "I can change it." Can I opt out of the feature that is trying to expand my AI usage? Can I delete the logs? Can I choose a less capable model that does not read my entire filesystem? Transparency without control is a window into a room you cannot leave.
The product teams that will win
Right now, most AI products treat transparency as a compliance checkbox. "We disclosed it in our terms of service" is not transparency. It is legal protection. Real transparency is visible, understandable, and actionable.
The product teams that will build lasting trust are the ones designing for all three layers.
- Visible provenance. Not a tiny link in a footer. A clear indicator on the content itself that says: this was made with AI, here is what that means, and here is what you can do about it. Google's ad labels are a start. Every product that surfaces AI-generated content should be asking the same question: does the person seeing this know what they are looking at?
- Explicit intent. When a feature is designed to expand usage, say so. Anthropic could have launched the same feature with a message like: "We think AI can help more people, so we are making it easier to try. Here is what it does, here is what it does not do, and here is how to turn it off." Instead, the framing was about convenience. The intent was about adoption. When those two things diverge and you do not acknowledge it, trust erodes.
- User control over data and behaviour. Transparency without control is surveillance with a nicer font. If I can see that your model was trained on my data, but I cannot remove my data or opt out, the transparency is a taunt, not a feature. If I can see that your feature is trying to make me use more AI, but I cannot adjust the level of suggestion, the disclosure is a disclaimer, not a choice.
My take
The AI industry is about to discover what every industry before it already knows: transparency is not what you are forced to show. It is what you choose to show when nobody is making you.
Google chose to label AI-generated ads before any regulator forced it to. That is a product decision, and it is the right one. OpenAI deleted logs during a copyright case. That is also a product decision, and it is the wrong one. Anthropic launched a feature to expand AI adoption and framed it as helpfulness. That is a product decision, and the jury is still out on whether it will cost them the trust they have spent years building.
The companies that win in AI are not going to be the ones with the best models. Models are being commoditised faster than anyone predicted. They are going to be the ones where the user can see what is happening, understand why, and change it if they want to.
Transparency is not a moat yet. But it is about to become one. The question for every product team is simple: are you building the moat, or are you hoping nobody notices you do not have one?