The series that translates national-security and AI-policy arguments across partisan lines, because the stakes are too high for tribal shorthand.
Congress Picks the Easiest Hard Question
What Happened This Week
On Tuesday, Reps. Jay Obernolte (R-CA) and Sara Jacobs (D-CA) reintroduced the Economy of the Future Commission Act of 2026, a bipartisan, bicameral commission to study how AI is reshaping the American economy and produce consensus policy recommendations to Congress. The same day, three additional bipartisan bills moved out of committee or got fresh floor scheduling: a hiring-algorithm disclosure bill from a House Education and Workforce subcommittee, a workplace-surveillance transparency framework cosponsored by Sens. Bob Casey (D-PA) and Bill Cassidy (R-LA), and a worker-data-portability proposal that picked up two new Republican cosponsors after weeks of being read as a Democratic-only effort. By Thursday, the AFL-CIO and the U.S. Chamber of Commerce had each issued statements that did not endorse any of the bills but did not oppose any of them, either. That is unusual.
Workforce AI is, suddenly, the only area of AI policy where Congress is moving with momentum. Frontier-model regulation is stalled. The federal preemption fight is in the executive branch and the courts. Defense AI is sitting in oversight hearings whose chairs cannot agree on what they are investigating. Workforce policy is the lane where the votes might actually be there. The question is what those votes are buying.
Here's What You Need to Know in 30 Seconds
The Obernolte–Jacobs commission would seat fifteen members across the House, Senate, executive branch, organized labor, and industry, with a 24-month mandate to produce a unified policy framework on AI's labor-market effects. The commission has no rulemaking authority. The three companion bills propose specific operational requirements: hiring-algorithm vendors would have to publish bias audits, employer surveillance products would have to disclose their data practices to workers in plain language, and workers would gain a portable record of the algorithmic data their employers had collected on them. None of these proposals breaks new conceptual ground. All of them are bipartisan. None of them tells you whether AI is going to take your job. That is the point of the commission, except the commission cannot answer that question either.
The Hawk Case: We Cannot Afford to Slow Down
The hawk position on workforce AI is, at root, that the United States is in a productivity race it cannot afford to lose. Automation is the single largest available source of total-factor productivity growth. Productivity growth is the only durable answer to demographic decline, public-debt service, and great-power competition with an adversary whose state-directed AI deployments are not constrained by labor-protection lawsuits. From this view, the Obernolte–Jacobs commission is a stalling tactic dressed up as deliberation. The commission will report in two years. In two years, the productivity gap will be measurable. In two years, labor-market dislocations the commission was supposed to anticipate will already have happened. The commission is not a forcing function. It is a way to be seen doing something while doing nothing.
The hawk goes further. The three companion bills, the hawk argues, will functionally raise the cost of deploying AI in domestic workplaces. Hiring-algorithm bias audits cost money. Surveillance disclosure requirements add compliance overhead. Worker-data portability creates a continuous obligation that follows a worker between employers, which means every employer pays for the data infrastructure even if the worker never uses it. None of these costs land on the productivity-leader companies — they land on small and mid-sized employers who cannot easily absorb compliance overhead. The result, the hawk says, is that the legislation does not slow AI deployment at the firms that matter; it slows it at the firms that already have less margin. Worse, it pushes employer surveillance into less regulated forms (third-party data brokers, contractors, gig platforms) rather than reducing the underlying practice. The legislation gestures at worker protection. It delivers compliance theater.
The deeper hawk concern is that the bipartisan posture itself is misleading. The fact that the Chamber of Commerce did not oppose this week's bills is read as institutional capture: industry has decided that bipartisan-looking workforce-AI bills are cheaper than the alternative, which is mandatory federal labor-AI rules with teeth. Industry wants the commission. The commission gives industry two years of regulatory predictability and a forum to shape the eventual recommendations. The commission's report, when it lands in 2028, will likely propose voluntary frameworks, public–private partnerships, and federal preemption of stricter state-level worker-protection laws. The bipartisan version is the version industry can live with. That is why it is bipartisan.
The hawk's bottom line: If Congress wants to address AI's effect on the workforce, the answer is to invest aggressively in worker retraining and education, accept that displacement is a real cost of a real productivity gain, and stop trying to legislate the deployment side of the equation. The commission's premise — that we need to study before we act — is the wrong one. We have been studying for forty years. Action looks like Pell Grants for AI literacy, expanded apprenticeships, and a tax code that rewards reskilling. Action does not look like algorithm audits.
The Reformer Case: Existing Labor Law Was Not Written for This
The reformer position starts from a different premise: the legal frameworks governing the American workplace pre-date algorithmic management by decades, and the gap is not closing on its own. The National Labor Relations Act was passed in 1935. The Fair Labor Standards Act in 1938. Title VII of the Civil Rights Act in 1964. These statutes were drafted in a world where employer decisions were made by human managers acting on legible criteria. They are now being applied, by judges and the EEOC, to decisions made by predictive models trained on data that the employer often does not understand and the worker has no access to. The reformer says: this is not a question of whether to regulate AI in the workplace. AI is already in the workplace. The question is whether the existing statutory floor still functions, and the evidence is that it does not.
Consider a concrete case the reformer points to. An algorithmic scheduling system cuts a worker's hours below the threshold at which the employer must contribute to her health insurance. Under existing law, the worker's recourse is a discrimination or retaliation claim — both of which require her to prove employer intent. There is no employer intent, in the conventional sense. There is a model. The model was trained on data the worker has never seen. The model is operated by a vendor the worker has never spoken to. The worker has no standing to sue the vendor and no theory of liability that fits the existing statutes. The 1935 statute did not anticipate this case. The reformer says: the statute now needs an update, and the update is not punitive. It is a baseline transparency floor. Workers should be able to see what data was used, what the algorithm decided, and what they can do about it.
The reformer extends this to the broader bipartisan package. Hiring-algorithm bias audits are not novel regulation. They are an extension of the disparate-impact framework that has applied to human hiring decisions for decades — the difference is just that the audit can now be conducted on the model rather than on the post-hoc outcomes. Worker-data portability is an extension of medical-record portability under HIPAA. Workplace surveillance disclosure is an extension of the consent frameworks that already apply to consumer data. None of these are radical. They are catch-up.
On the commission specifically, the reformer is more sympathetic than the hawk's framing suggests. The reformer reads Obernolte–Jacobs not as a stall but as a forcing function. A bipartisan commission with a 24-month mandate produces a report. The report becomes a Schelling point. Even if the commission's recommendations are not implemented in full, the act of producing the report changes what counts as the default policy conversation. The reformer notes, correctly, that nearly every major piece of post-1990 economic policy passed through a commission stage first. The 9/11 Commission. The Greenspan Commission on Social Security. Simpson–Bowles. Commissions get a bad rap from people who do not have to legislate. People who do legislate know that a commission is sometimes the difference between a bill that passes and a bill that does not.
The reformer's bottom line: The existing legal floor for the American workplace is fraying under algorithmic management, and Congress is finally moving on the most defensible piece of the problem. The hiring-algorithm and surveillance bills are not theater — they are the minimum viable update to civil-rights and consumer-protection frameworks for the AI era. The commission is not a stall — it is the political infrastructure that makes the next round of legislation possible. The hawk's preferred response (retraining only) is a half-measure. Retraining matters, and Congress should fund it. But retraining without a baseline rule on algorithmic management is the worker bearing the entire cost of a transition the employer is unilaterally driving.
Where They Actually Agree
Strip the framing away and the hawk and the reformer agree on more than either side will say in public.
Both agree the workforce dislocation is real. Both agree the existing statutory framework is not fit for purpose. Both agree that algorithmic transparency, in some form, is going to happen — the disagreement is over how mandatory and how soon. Both agree that the federal government is going to spend a lot of money on retraining over the next decade, and both agree that the existing apprenticeship and Workforce Innovation and Opportunity Act infrastructure is the wrong delivery mechanism. Both agree that the commission's report, if produced, will be cited regardless of which party controls Congress at the time of its release. And both agree, privately, that the bipartisan posture this week is the easiest version of a much harder conversation that will start in earnest the moment AI displacement becomes a frontline campaign issue.
The hawk and the reformer also agree that this is the lane Congress can move in because it is the easiest hard question. Frontier-model regulation requires Congress to define what a frontier model is, which requires judgments about capability the major labs have lobbied to keep ambiguous. Federal-state preemption requires Congress to override state legislatures, which is unpopular even with members who would otherwise support a national framework. Defense AI requires Congress to legislate against the Pentagon's procurement process, which has its own political constituency. Workforce AI is none of these things. It is a question about the workplace, where Congress has been writing rules since 1935, where the relevant statutes already exist, and where the operational asks are incremental rather than transformative. So this is the lane that moves.
Where They Don't (And Shouldn't Pretend To)
On whether algorithmic management is a different kind of relationship from human management. The hawk treats algorithmic decision-making as a more efficient version of the same human function: better, faster, less biased, no real change to the workplace contract. The reformer treats it as a different kind of relationship altogether, opaque and asymmetric, operating on data the worker has no access to. They are using different working models of what an employer is. This disagreement does not get resolved by data. It gets resolved by policy choice.
On the cost incidence of compliance. The hawk argues that compliance cost lands disproportionately on small and mid-sized employers and pushes practices into less-regulated forms. The reformer argues that compliance cost is a feature, not a bug — it forces employers to internalize costs they currently externalize onto workers and the broader labor market. These are both empirically defensible. The relevant data on which one dominates does not yet exist, in part because the legislation that would generate it has not passed.
On commissions as a tool of governance. The hawk reads the commission as institutional capture. The reformer reads it as institutional infrastructure. The history of commissions in U.S. economic policy is mixed enough to support both readings. The commission's value depends almost entirely on staffing, which Obernolte–Jacobs leaves unspecified.
On retraining alone as the answer. The hawk treats retraining as the load-bearing intervention. The reformer treats retraining as one piece of a larger framework. The Section 32 of the Workforce Innovation and Opportunity Act apprenticeship data, which both sides cite, supports neither claim cleanly. Retraining works when displaced workers have time to retrain, infrastructure to retrain in, and a reasonable chance of equivalent employment afterward. Whether AI displacement satisfies those preconditions is the question both sides assume the answer to.
Here's My Two Cents
Workforce is the easy hard question for Congress, and I think Congress is going to pass at least two of these four bills before the recess. That is genuinely good news, and it is also a problem.
The good news first. Bipartisan workforce-AI legislation is the first federal AI policy this Congress has been able to advance. The fact that the AFL-CIO and the Chamber issued non-opposing statements is, in legislative terms, an alignment of stars. Worker-data portability and hiring-algorithm disclosure are the kind of incremental, statute-extending updates that the existing legal apparatus can absorb without rewriting itself. The commission is, at minimum, a venue where the labor-market data people keep promising and never producing might actually get produced. None of this is theater. All of it is real.
The problem is what it implies about everything Congress is not doing. Workforce AI is moving because it is the easiest hard question. That is the same reason it is not where the most consequential decisions are being made. The federal preemption fight, the frontier-model definition fight, the defense-AI procurement fight, the AI-and-elections fight — those are where the policy weight is. None of them are moving. The fact that workforce can move and the others cannot tells you something about the political economy of AI legislation. It tells you that bipartisan posture is available exactly when the legislative ask is small enough to be uncontroversial. It is not available at the scale of the actual problem.
I think the Obernolte–Jacobs commission is the right move on the merits. I also think the commission is going to be undermined, in slow motion, by exactly the dynamic that made it possible. The 24-month mandate is going to overlap with at least one campaign cycle. The commission's interim findings are going to leak. Those leaks are going to be selectively cited by both sides, in ways that harden positions before the final report can do its job. By the time the report lands in 2028, the policy conversation it was meant to shape will have moved on. This is what happens to commissions in an information environment that does not give them room to deliberate. The 9/11 Commission worked partly because it operated in an environment where the public was prepared to give it the space to produce findings. Obernolte–Jacobs will not have that room. Almost no commission has, since.
That makes the three companion bills more important than they look. If the bills pass, they create operational rules that exist in statute regardless of whether the commission produces a coherent report. The bills are the floor. The commission is the ceiling. The bills do not require the commission to succeed. The commission needs the bills to pass — because if the commission's report lands without any companion legislation in force, it lands as a recommendation document with nothing already moving in its direction, and recommendation documents that arrive into a vacuum tend to stay there.
The hawk's framing has the politics backward. The bipartisan workforce package is not a stall. It is the only AI legislation that has a path to becoming law before 2028. Killing it on the grounds that it is incremental is not a posture for someone who wants stronger action. It is a posture for someone who wants no action at all. If the hawk genuinely believes retraining is the right answer, the package is the version of retraining that includes the data infrastructure to know whether retraining is working. Without the package, Congress will fund retraining and have no idea whether the funding is reaching the right workers.
The reformer's framing has its own problem. The reformer is right that the existing statutory floor needs updating. The reformer is also a little too sanguine about the commission's value. A commission, in the present information environment, is a venue where staff write a report under pressure from constituencies that have already decided what they want the report to say. The reformer should be advocating for the bills first and the commission second, not the other way around. The bills are the durable artifact. The commission is the political theater that makes the bills possible. Both are necessary. They are not equally important.
The deeper observation, the one I keep returning to, is that workforce AI is the lane where Congress is operating at all because it is the lane where the existing legal framework gives Congress something to work with. The 1935 NLRA gives Congress a baseline. The 1964 Civil Rights Act gives Congress a baseline. The 1996 Telecommunications Act gives Congress a baseline. AI in critical infrastructure has no baseline. AI in defense procurement has no baseline. AI in elections has no baseline. The reason workforce moves is not that workforce is more urgent. It is that the political technology for legislating workforce already exists. The political technology for legislating frontier-model deployment, military AI, or election integrity does not exist yet, and Congress has not been willing to build it.
That is the structural fact this week's bills illuminate. Congress is moving where the existing law gives it permission to move. Where the existing law does not give Congress permission, Congress is not moving. The question that will define the next two years of AI policy is whether Congress can build the political technology to legislate in the lanes where existing law is silent. The Obernolte–Jacobs commission is not going to build that technology. The bills, if they pass, will not build it either. They will, at most, hold the line on the workforce. And holding the line on the workforce, while everything else gets decided in the executive branch and the federal courts, is not a complete answer. It is the answer Congress can give right now. We should take it. We should not pretend it is enough.
Related Briefings
Anna R. Dudley writes on national security, intelligence policy, and the places where hawks and reformers need to find each other. Bipartisan Translation is the weekly series for the conversation that is not happening on cable news. Subscribe at annardudley.substack.com.