Amazon has said it plans to stop using nondisclosure agreements (NDAs) in talks with local governments about data center construction. Meanwhile, startups developing consumer AI agents are aiming to offer services that users will grant access to their email, files and credit cards.
Amazon Revises Confidentiality in Talks With Local Governments
An NDA is an agreement not to disclose information learned during negotiations, among other things. Amazon’s new policy applies to talks with local governments about data centers; it does not mean the company is abandoning all NDAs. Microsoft is also said to have taken a similar step in the first half of 2026, according to documents that were made public.
The backdrop is public opposition to the infrastructure needed to support AI. Secrecy in negotiations has fueled distrust, and the decision not to use NDAs can be seen as removing one barrier to local communities understanding the plans.
But ending confidentiality obligations is not the same as proactively disclosing necessary information. The documents do not say what will be disclosed, when or to what extent. It would be premature to conclude that a policy change alone will win residents’ understanding.
Opposition to Construction Is Spreading to Project Freeze Proposals
Opposition to AI infrastructure has gone beyond voicing concerns. The documents say that hundreds of proposed or enacted moratoriums on development and other activities have been reported from New York to San Francisco.
That figure includes both proposals and measures that have already taken effect. It does not mean that construction has been halted in hundreds of communities. The documents also do not specify the scope or duration of the measures in each area, or the specific reasons for opposition.
The mechanisms that make it easier for companies to negotiate do not necessarily make it easier for residents to assess plans. To evaluate Amazon’s NDA policy change, it is necessary to look beyond whether agreements are used and determine whether residents actually gain access to more information they can use to make decisions.
Consumer AI Raises Questions About Trusting Services With Personal Information
While disclosure is the challenge for local communities, consumer AI raises the question of how much access users are willing to grant. AI agents are software designed not merely to provide answers but to carry out tasks on users’ behalf.
Access to email, files and credit cards could expand the range of tasks an agent can perform. In turn, users must decide how much authority to grant over their private information and payments. However, the documents provide no figures showing that such services have become widely adopted by consumers.
Tab, Underdog and Hark are presented as companies that emphasize privacy. How a service handles the information entrusted to it will be a consideration alongside convenience. The program also raises the possibility that, depending on the business model, an agent meant to act on users’ behalf could instead play a role in promoting sales.
A User’s Permission Does Not Guarantee a Website Will Accept an Agent
Consumer AI agents face another hurdle, beyond earning users’ trust. Even if users want to delegate tasks, the range of actions an agent can perform will be limited if the websites it interacts with do not accept third-party agents.
Two kinds of permission need to be considered separately. One is the user’s permission for AI to access their information or payment-related capabilities. The other is the service provider’s permission for that AI to access or operate its service. Securing only the former does not guarantee that the service can function.
The documents also discuss barriers to access for external agents at United, Apple and Amazon. However, they do not specify the restrictions, the actions they apply to or the reasons behind each company’s policies.
Comments