Could AI assistants become gatekeepers of what people see?

The report, from the Ada Lovelace Institute (Harry Farmer et al.), analyses "the biggest policy challenges presented by Advanced AI Assistants" — AI apps or integrations that "engage in fluid, natural-language conversation," show "high degrees of user personalisation," and "adopt human-like roles in relation to their users." Its central structural claim is that "Assistants could become the principal means by which most people access the internet and interact with the digital world – as universal digital intermediaries – giving them substantial gatekeeping power over users." The report calls this "a model of computer use that would place Assistants in a position of subtle but enormous power," able to "determine what information, viewpoints, ideas and options people are exposed to; how these are presented and framed; and how user orders and requests are interpreted."

The reasoning: Assistants are "typically easy to use, highly personalised and 'personable'," which the report says makes them "very persuasive and easy to trust"; combined with aggressive marketing and integration "into existing apps, productivity suites and operating systems," they can "spread rapidly throughout our economies and societies" and reach the scale needed to become a universal layer. From that single structural position the report derives six distinct harms that "could arise even if we make the very optimistic assumption that Assistants function exactly as advertised": failed economic benefit, privacy and security exposure, market distortion and monopoly power, hard-to-detect political influence, "widespread cognitive and practical deskilling," and harm to "mental health and flourishing."

This sits alongside What makes ethics of AI assistants fundamentally different from chatbots?, which maps the same category of system — action-taking, natural-language-interface agents — across individual and societal ethical rings; the Ada Lovelace report narrows that terrain to one policy mechanism, gatekeeping power from intermediary scale, and frames it as a diffusion-management problem rather than a design-ethics one. Its persuasion claim ("very persuasive and easy to trust") matches the mechanism in Does chatbot language style actually shape how much we trust it?, where trust is produced by linguistic signals rather than verified expertise. Its mental-health risk is consistent with, but broader than, the measured finding in Does sustained engagement with AI companions harm well-being?, which traces one specific pathway the report leaves unspecified. And its deskilling and dependency claim is a more general, anticipatory version of the pattern quantified in Where have workers actually delegated tasks to AI?, which measures actual delegation rather than asserting its future scale.

The excerpt is a policy analysis, not a measurement: none of the six harms is evidenced here with data, and the report's own "gatekeeping power" claim is conditional on Assistants reaching the scale and capability "anticipated by the tech industry" — a premise rather than an observed fact. What the excerpt supports, at the strength available, is only that the six risks are logically downstream of one condition, intermediary-scale adoption; the governance question the report is actually posing is how to shape or forestall that condition, not how to mitigate each harm independently.