Signal
The AI a company buys increasingly runs on other companies' AI, and buyers are starting to ask to see the chain. An audit of 2,400 business software providers that advertise AI found that most do not name the other AI providers working behind them. Last year one website chat agent showed what that hides: a single integration let attackers into the customer records of hundreds of organisations, Cloudflare among them. The response is arriving as plumbing and rules. The internet's standards body has chartered a group so that an agent can prove which company operates it, the Linux Foundation now hosts directories where agents from different vendors list themselves, and Europe's AI law will require written agreements between a high-risk AI provider and every supplier of the models, tools and components inside it. Meanwhile the UK wants a know-your-agent standard for payments and Singapore has rewritten its guidance for the case where one agent hands work to another. The question a buyer used to ask was which vendor. It is becoming which vendors, and how far down.
So what?
The AI you buy is a chain of companies you have never met, and knowing who is in it is becoming part of the deal.
Evidence
IAPP, 22 July 2026 (study): DataGrail reviewed the data protection assessments of 2,400 business software providers that advertise AI features and found that 63.6 percent fail to disclose that another AI provider processes data behind them. Its chief executive said a customer may be told about one subprocessor while the vendor uses two, three or four.
HM Treasury, 14 July 2026 (policy): the UK's financial services AI adoption plan, whose recommendations the Treasury accepted, asks for a trust framework for payments made by AI agents: Know Your Agent protocols with standard identity and verification for autonomous software agents, and clear liability where multiple agents and third parties operate across complex value chains. The UK has no registry of AI agent operators yet.
IMDA, 20 May 2026 (policy): Singapore updated its governance framework for AI agents after feedback from over 60 organisations, among them AWS, DBS, Google and Salesforce, adding guidance on the risks of multi-agent systems and third-party agents. Its case studies include a government agency that kept external tools out of the first rollout of its coding agents until logging and monitoring were in place.
Worth asking
Think of the software you rely on most at work. If it ran on three other companies' AI behind the scenes, what would you want to know about them?
