The Language Access for All Act Just Settled an Argument Our Field Has Been Having for Years

For the past few years, every conference panel on the future of interpretation has circled the same question: what happens to human interpreters when the machines get good enough? I have sat on those panels. I have watched colleagues brace for a slow decline into irrelevance. And I have watched procurement teams quietly test whether a chatbot could replace a certified medical interpreter in an exam room.

Congress just answered the question, and the answer is no.

What the bill actually does

In March 2025, Executive Order 13166, the 25-year-old federal mandate for language access, was revoked. Civil rights advocates and members of Congress responded with a legislative push to restore that mandate permanently rather than leave it dependent on whoever occupies the White House. The House version, H.R. 7223, was introduced in January 2026. The Senate companion, S. 4985, followed in July and is now moving through the Judiciary Committee.

The Language Access for All Act requires covered federal agencies to publish a documented language access plan, certify compliance annually to the Attorney General, and provide meaningful access to LEP individuals as a legal obligation rather than a courtesy. None of that should surprise anyone who has spent time in this field. Title VI has required something close to this for decades. What is new, and what should get every interpreter's attention, is the provision that draws a hard line around AI.

The bill prohibits agencies from fully replacing qualified human translators or interpreters with AI or machine translation. Not as guidance. As statute. Agencies "shall require a qualified human translator or interpreter" for the services this bill covers. Machine translation post-editing, where an AI-generated draft goes through a qualified human reviewer, remains permitted. Submitting AI output as official communication without that review does not.

Why this matters even if you never work with a federal agency

If you are reading this as a working interpreter, your first instinct might be that a federal statute doesn't touch your caseload. Look closer. Title VI already binds any organization that receives federal funding, which means most hospitals, courts, universities, and social service agencies in the country. The Joint Commission's Accreditation 360 framework, effective this January, made language access a formal patient safety requirement under National Performance Goals 4 and 7, and it explicitly disqualifies untrained bilingual staff and family members from serving as interpreters. San Francisco's updated Language Access Ordinance is expanding required languages based on demographic shifts. The 2027 deadlines for ADA Title II and HHS Section 504 compliance are approaching for state and local agencies.

None of these bodies are waiting for AI to mature into something trustworthy enough to hand a clinical consent form to. They are writing rules that assume it can't be, at least not without a qualified human in the loop.

AI stopped being the differentiator a while ago

Here is the part I want my colleagues in this field to sit with. For a stretch of time, the industry conversation treated AI adoption as the thing that separated forward-looking language service providers from the ones getting left behind. That framing is out of date. Every serious provider now has some form of AI-assisted translation in its workflow. McKinsey's research on this points to the same shift: the organizations pulling ahead aren't the ones using AI, they're the ones with defined processes for deciding when a model's output needs a qualified human to validate it, especially in legal, medical, and public-facing content.

That is the differentiator now. Not whether you use AI. Whether you can tell a client, a court, or a surveyor exactly where the human judgment sits in your process, and prove it.

This is good news for interpreters, but it is not a free pass. The Language Access for All Act, the Joint Commission's standards, and Title VI enforcement all point toward the same expectation: documented, qualified, accountable human involvement. A certification you can produce. A quality assurance process you can describe. A named professional who stands behind the accuracy of what was said or written. AI cannot supply any of that, and increasingly, the law says it isn't allowed to try.

What this means for how we position ourselves

I built EITLA, our governance framework for language access, around exactly this distinction, because I watched too many organizations confuse having a translation tool with having a defensible language access program. The two are not the same thing, and the gap between them is where accreditation findings, civil rights complaints, and preventable harm live.

If your organization is still treating your AI workflow as a finished answer rather than the first step in a process that ends with a qualified human, this bill is your signal to close that gap before a surveyor or a plaintiff's attorney finds it for you. I cover this in more depth on our Language Access Matters podcast, and our EITLA program walks compliance and language services teams through building the documentation this moment requires. If you want the shorter version first, our biweekly newsletter tracks developments like this one as they happen.

The machines are not going away, and they shouldn't. But the law just confirmed what most of us already knew from the exam room and the courtroom: judgment, accountability, and qualification are not features you can automate. They are the job.

Carol G. Velandia is the founder and CEO of Equal Access Language Services and the creator of the EITLA framework for language access governance.

‍