TThe Diary of a CEO
← All frameworks
Influence

AI Accountability Leverage Map

Find where AI expansion depends on you, then contest or redesign that dependency

Difficulty
Advanced
Time to result
~months to results
Steps
7
Confidence
93%

The AI Accountability Leverage Map begins from Hao's claim that AI expansion is not inevitable or frictionless. Companies still require data, labor, physical infrastructure, institutional adoption, and favorable rules. Map where your life or organization intersects with those dependencies: supplying content, hosting a data center, setting a school or workplace adoption policy, buying a system, or shaping local regulation. Identify who controls each decision and organize with others who share the impact. Set a concrete demand such as consent, compensation, transparency, environmental limits, or a different procurement choice. Use the legitimate channel available, including institutional policy, public comment, collective organizing, or legal action with appropriate counsel. Pair opposition with support for a more efficient or accountable alternative. The mechanism converts diffuse anxiety into a bounded contest over an identifiable dependency and measurable decision.

Origin

Hao asks listeners to examine every way their lives intersect with the resources and deployment spaces AI companies need, then use those points to slow harmful practices and build alternatives.

Core principles

  • 01Large AI companies still depend on data, infrastructure, adoption, and permission
  • 02Individuals gain leverage by acting through shared institutions and affected groups
  • 03The goal is accountable development rather than eliminating all AI
  • 04Resistance and alternative development can proceed together
  • 05Local action can contest a global industry's expansion

How to run it

  1. 1

    Map the dependency

    List the data, labor, land, utilities, purchasing, or institutional access that an AI project needs from you or your community. Select one intersection where a real decision is still open.

    Pro tip Look beyond product use to schools, workplaces, local planning, and creative work.

    Watch out A broad concern without a decision point is difficult to act on.

  2. 2

    Locate authority

    Identify the person, board, agency, court, or procurement process that can change the decision. Confirm the relevant deadlines and procedures.

    Watch out Do not assume a technology company is the only actor with decision power.

  3. 3

    Build the affected group

    Connect with workers, creators, parents, residents, or institutional members who share the impact. Gather their direct evidence and priorities.

    Pro tip Let affected people define the harm and desired remedy in their own terms.

  4. 4

    Set a concrete demand

    Ask for a specific change in consent, compensation, transparency, resource limits, procurement, or deployment. Define what acceptance would look like.

    Pro tip Target the exchange or governance mechanism rather than arguing that all AI has no utility.

    Watch out Legal claims and high-risk disputes require qualified advice.

  5. 5

    Use the available channel

    Act through the relevant policy meeting, institutional process, collective campaign, public record, or lawful legal route. Coordinate evidence and messaging around the chosen demand.

    Watch out Do not expose private information or vulnerable people without consent.

  6. 6

    Build the alternative

    Where useful technology is still needed, support a narrower, more efficient, or more accountable approach. Show that contesting one production model is not the same as rejecting every capability.

    Pro tip Use the Bicycles Versus Rockets test to right-size the alternative.

  7. 7

    Measure the decision

    Track whether policy, construction, adoption, compensation, transparency, or system design changes. Revise the leverage map when authority or dependencies shift.

    Watch out Attention without a changed decision is not yet the intended outcome.

In the wild

Communities contesting data centers

Hao says protests against data centers have occurred in the United States and elsewhere, with some projects reportedly stalled and some localities banning development. The episode does not independently enumerate those cases, but she uses them to argue that local infrastructure decisions remain contestable.

A physical dependency creates a local decision point for collective action.

Creators contesting training-data use

Hao points to artists and writers suing AI companies over intellectual-property use. She presents these cases as attempts to create mechanisms for withholding data or changing the terms under which creative work can be used.

A diffuse concern about training data becomes a concrete dispute over rights and exchange.

Common mistakes

Assuming expansion is automatic

Hao argues that companies still need data, infrastructure, adoption, and favorable decisions, each of which can become a leverage point.

Demanding only a total stop

Her stated objective is to end imperial practices while preserving utility through more efficient and broadly beneficial alternatives.

Acting alone on a systemic issue

The examples in the episode rely on communities, professional groups, institutions, or multiple plaintiffs rather than isolated consumer preference.

Is it for you?

Best for

Creators, workers, schools, companies, and communities deciding how to respond to an AI deployment or infrastructure proposal.

Not ideal for

Urgent personal safety, employment, or legal disputes that require qualified professional support rather than a general action map.

From the transcript

think about all of the ways that your life intersects with the resources and the that the AI industry needs

Karen Hao · (2:02:30)

let's not make it go flawlessly if we don't agree with what they are doing

Karen Hao · (2:04:00)

let's break up the empire and let's forge new paths of AI development

Karen Hao · (2:04:30)

From the episode

AI Whistleblower: We Are Being Gaslit By The AI Companies! They’re Hiding The Truth About AI