Mozilla has helped launch a map of 79 venture-backed companies building what it calls trustworthy AI infrastructure. They are sorted across seven layers of the AI stack, among them identity, data control, interoperability and evaluation. To be included, trust must be “baked into its product or business model through human agency, trust by design or aligned business models”. Trust appears here as work that has to be designed, funded and maintained.
The phrase is easily emptied of meaning. Every company wants to be trusted, especially when its product is new enough to make customers uneasy. But a system earns trust through more ordinary matters: who controls the data, whether its claims can be tested, whether decisions can be traced, and whether a customer can leave without losing everything.
I know lock-in from the inside. For almost two decades, when I was developing enterprise applications delivered through a browser, compatibility with Internet Explorer 6 was non-negotiable, because so many in-house systems depended on proprietary extensions only it supported. Even after Microsoft stopped supporting it, we kept catering for its quirks, and the extra development and support cost was passed on to our customers. Mozilla’s own Firefox helped to break that grip. Nobody will ever add up the global bill, but I know it was enormous.
That is why the business-model test matters more than the badge. Mozilla’s own pitch ends: “trust is not just a virtue; it’s a path towards profitability.” Perhaps. But a company whose revenue depends on extracting more data, locking in customers or making oversight inconvenient will eventually find that trust and profit point in different directions, and that its assurances have limits. Good intentions remain useful, but they need a structure capable of carrying them.
Prompted by “This map shows the most important asset that VC-backed companies need right now” on the Mozilla blog: https://blog.mozilla.org/en/mozilla/ai/trustworthy-ai-map/