Senator Markey's AI Accountability Bills Take Aim at Hiring Algorithms, Data Centre Waste, and Worker Surveillance

Senator Ed Markey, Democrat of Massachusetts, introduced a package of AI accountability bills in July 2025, targeting the everyday harms that large-scale artificial intelligence systems already cause across American workplaces and communities.

What the Bills Would Actually Do

Markey's package covers four distinct problem areas. Each one reflects a real failure happening now, not a hypothetical future risk.

Automated Hiring Tools

Most large employers now use software to screen job applications before any human reads them. An algorithm, a set of programmed rules a computer applies to reach a decision, scores candidates and can reject them without a recruiter ever seeing their name. Markey's hiring-focused bill would require companies to tell applicants when such tools are in use. It would also give people a formal mechanism to challenge a rejection that came from software rather than a person.

This matters because the stakes are high and the transparency is near zero. A 2023 study cited by the AI Now Institute found that automated screening tools trained on historical hiring data routinely penalise candidates from demographic groups that were underrepresented in past successful hires. The pattern is not a bug. It is baked in.

Data Centre Environmental Costs

A data centre is, at its core, a building full of servers running constantly. AI workloads are among the most power-intensive computing tasks ever designed. They also require enormous quantities of water for cooling. Neither cost appears cleanly on any public ledger today.

Markey's proposal would change that by mandating disclosure of electricity consumption and water usage at the facility level. The argument is simple: if a company is burning municipal water supplies to generate profit, the public deserves to know the quantity. The legislation would put those numbers somewhere visible rather than buried in sustainability appendices that few people read.

Workplace Surveillance

The third strand of the package addresses how employers use AI to monitor workers. Warehouse workers report systems that track their movement rates in real time. Call centre employees describe software that analyses the emotional tone of their voice. Even office workers face tools that log keystrokes or measure how long they spend away from their screens.

Markey's bills would restrict what data employers can collect through these means and place limits on how it can be used against workers. The principle is that surveillance of this granularity creates a power imbalance that existing labour law was never designed to handle.

Algorithmic Bias

The fourth target is algorithmic bias in broader contexts, including financial services and consumer decisions. Software trained on historical data inherits the biases embedded in that history. When those patterns drive decisions about who gets a loan, who gets shown a job posting, or who gets flagged for fraud review, the discrimination is real even if no individual person made a biased choice.

The Economic Inequality Argument

Markey introduced the package at a moment when a small number of technology companies are reporting record profits from AI products while wage growth in many sectors that AI directly affects has stalled. The bills do not attempt to ban AI-generated profit. They push back against an arrangement where the financial gains flow exclusively toward shareholders while the costs, environmental, social, and economic, land on everyone else.

One statistic frames the problem sharply. The International Energy Agency projected in its 2024 report that data centre electricity demand worldwide could double by 2026. That growth is driven primarily by AI. The communities near those facilities, and the ratepayers on the same grid, absorb that cost with no compensation and little information.

"We cannot afford to let Silicon Valley write the rules for a technology that affects every American worker and every American community," Markey said in remarks accompanying the introduction of the bills.

Which Controls Failed, and What Defenders Should Learn

This is not a cybersecurity breach in the traditional sense. No attacker exploited a misconfigured firewall. But the failure modes Markey's bills address are deeply familiar to anyone in the security and compliance space: absent oversight mechanisms, opaque decision-making systems, and the assumption that if something is technically legal it is therefore acceptable.

Automatic hiring tools represent a human-process failure at scale. When organisations replace human judgment with algorithmic screening and then remove the audit trail that would allow anyone to challenge the output, they create a black box with real consequences. The same dynamic appears in security contexts when automated threat-detection systems make access decisions without logging or appeal mechanisms. The lesson is identical in both cases: any automated system that affects a person's livelihood or access rights needs a human-readable audit log and a formal challenge process.

The data centre transparency gap is a governance failure. Security teams understand this pattern well. Untracked assets create unmanaged risk. When the resource being consumed is water or electricity rather than bandwidth or compute credits, the accountability principle does not change. What gets measured gets managed. What stays invisible does not.

Organisations that want to get ahead of incoming AI regulation should start with an internal audit of every automated decision system they operate, including HR tools, customer-scoring platforms, and any software that routes or filters human requests. That audit should answer three questions: what data does the system use, who can challenge its outputs, and what is the environmental cost of running it. If your team cannot answer all three, the gap is the risk.

Building that kind of policy-aware, critically literate workforce is exactly the kind of challenge that security-awareness training is designed to address, because the human layer is where both AI misuse and regulatory non-compliance most often take root.

The legislative window for sensible rules is narrow. Organisations that treat AI governance as a compliance checkbox rather than a genuine operational discipline will find themselves on the wrong side of whatever passes. Review your AI vendor contracts now. Confirm whether any tool that influences hiring, performance management, or customer decisions carries a bias audit. If it does not, ask the vendor why.

For a practical starting point on mapping these gaps against established frameworks, the Train2Secure standards library covers NIST AI RMF and related controls.