Skip to content

Region Focus | United States

Nov 6, 20261 min
TL;DRPentagon Chief Technology Officer Emil Michael said on 11/4 that the federal government will publish a complete inventory of laws applying to domestic AI labs later this month. OpenAI policy chief Chris Lehane is preparing to ramp up Washington lobbying once Congress's lame-duck session opens on 11/9, pushing mandatory 'speed limits' for frontier AI. Microsoft's National AI Safety & Privacy Standard, negotiated with the American Federation of Teachers (AFT), took contractual effect across US school districts on 11/1, banning the use of student data to train models and requiring human oversight for high-stakes decisions.

Look up & save

Loading dictionary…

Region: United States

The individual moves of big US labs — Anthropic, OpenAI, Google and the rest — already show up daily in this site's regular coverage. What's worth a region-level write-up this week is something different: three completely separate rule-making channels moving at once. A federal agency is systematizing existing law, OpenAI is trying to define what "speed limits" mean before Congress reconvenes, and Microsoft is using contract leverage to make a standard take effect in public schools without waiting for any legislature. What ties the three together isn't a debate over whether to regulate AI — it's a contest over who gets to write the rules, and through which channel.

Key Developments This Week

Pentagon to publish a full inventory of laws applying to AI labs this month

Pentagon Chief Technology Officer and Super Intelligence Task Force vice chair Emil Michael said on 11/4 that the federal government will publish, later this month, a complete inventory of existing laws and regulations that apply to domestic AI labs. According to reporting, this isn't new legislation — it's a mapping of current federal statutes onto AI development workflows. For labs under contract with the government, including those working on classified projects, the inventory will offer concrete guidance on data security, model transparency, and accountability standards, aiming to give commercial AI companies and defense agencies legal certainty for their collaboration rather than waiting on Congress to pass something new. (Reuters · CDO Magazine · emergent.sh)

OpenAI gears up for a Washington lobbying push as Congress's lame-duck session opens

OpenAI chief global affairs officer Chris Lehane told Axios the company has spent months assembling a slate of policy proposals — some new, some repackaged — ahead of a full lobbying push once Congress's lame-duck session (the stretch between the election and the new Congress being sworn in on 1/3) begins on 11/9. The central ask is mandatory safety rules for the most powerful AI systems, which OpenAI is calling "speed limits," framed around keeping humans in control of the technology. Lehane also plans to rebrand data centers as "AI Compute Campuses," emphasizing economic benefits for host communities in response to growing local backlash against data-center buildouts. He said the campaign will run through the 2028 presidential race, targeting potential candidates to make AI a central policy issue. (Axios · the-decoder · BigGo Finance)

Microsoft's school AI standard, negotiated with the teachers' union, takes effect

The National AI Safety & Privacy Standard — announced jointly on 9/9 by Microsoft vice chair and president Brad Smith, American Federation of Teachers (AFT) president Randi Weingarten, and United Federation of Teachers (UFT) president Michael Mulgrew — became available to every US school district on 11/1. Districts don't need to renegotiate; they simply add the terms to a new or existing Microsoft contract, which turns the privacy commitments into a legally enforceable contractual obligation. The standard includes 10 enforceable protections: no use of student or educator data to train models (barring a narrow safety exception), no sale of data or use for advertising, a ban on AI companion features designed to foster emotional attachment or extend engagement, a ban on student monitoring, human oversight for high-stakes decisions, and a 72-hour breach-notification requirement. Districts that find a violation can terminate the contract and seek damages. (Microsoft Education official announcement · AI Chat Daily)

This week's sequence of US AI regulation and lobbying news
This week's sequence of US AI regulation and lobbying newsMicrosoft and AFT's school AI standard took effect 11/1; the Pentagon previewed its AI-lab regulatory inventory on 11/4; Congress's lame-duck session opened 11/9, with OpenAI launching its Washington lobbying push. 11/1 School AI standard live Microsoft × AFT contract terms 11/4 Pentagon previews inventory Full AI-lab law list due this month 11/9 Lame-duck session opens OpenAI's "speed limits" push begins
  1. 11/1 — Microsoft and AFT's school AI standard takes contractual effect, banning training on student data
  2. 11/4 — Pentagon CTO Emil Michael announces a full AI-lab regulatory inventory coming this month
  3. 11/9 — Congress's lame-duck session opens; OpenAI launches its lobbying push for federal "speed limits"

X-axis is proportional to actual elapsed days. 11/1 = T+0, 11/4 = T+3 days, 11/9 = T+8 days. Scale k=(690-70)/8=77.5 units per day, x=70+days×77.5.

Deep Dive

I'd read this week's biggest US signal through institutional theory's concept of coercive isomorphism — three actors, each reaching for a different institutional channel to decide how AI agents get governed, in the same regulatory vacuum.

A federal agency's coercive isomorphism. The Pentagon's inventory doesn't create new obligations; it translates existing law into a compliance map AI labs can actually read, reducing uncertainty rather than tightening the regulatory bar. That's good news for vendors already doing government contract work — the rules just got clearer — but it also raises the bar for new entrants without the legal resources to parse the inventory in the first place.

A company lobbying to shape the coercive isomorphism before it forms. OpenAI's "speed limits" pitch is, at bottom, an attempt to get its own proposed language adopted as industry-standard vocabulary before Congress actually legislates anything — getting ahead of binding rules by shaping their content first. Notably, Lehane is mapping this campaign all the way through the 2028 election, which signals OpenAI is treating this as a long political project, not a one-off lobbying push.

A private contract bypassing legislation entirely to manufacture a de facto standard. What's most distinctive about the Microsoft-AFT standard is that it required zero government legislation. Using Microsoft's contract leverage in the K-12 market, union-negotiated terms became an add-on every US school district can simply opt into — a legally enforceable obligation that bypasses Congress and every state legislature. And it moved faster than any federal or state bill could: announced 9/9, effective 11/1.

The common thread across all three is that, in the vacuum left by the absence of unified federal AI legislation, different actors have each found their own way around Congress — an agency through administrative interpretation of existing law, a company through lobbying aimed at shaping future legislation, and a major vendor through contract leverage that manufactures a de facto standard outright. That's a different kind of fragmentation than the "state-law patchwork plus capital concentration" pattern observed in earlier weeks' US coverage: this time the split isn't geographic (states each setting their own rules) — it's a split in the institutional channels themselves, over who gets to use which lever to write rules at all.

Takeaways for Taiwanese Founders

  • If you build AI/agent systems for government or defense contracts: once published, the Pentagon's inventory will become the first checklist for assessing whether you can bid on US government AI work. Teams with real security and systems-integration experience should start mapping their own compliance gaps against the inventory's likely scope now, rather than waiting for a specific RFP to force the issue
  • If your customers are institutions with a powerful professional union or buyer coalition — education, healthcare, and similar sectors: the Microsoft-AFT contract model shows a path to a binding standard that doesn't require waiting on government legislation. If your agent product sells into Taiwanese schools, hospitals, or public agencies, consider negotiating contract-level protections directly with the buyer's union or association, rather than waiting for a regulator to issue guidance — you may earn trust faster than competitors who wait
  • If you're weighing whether to adopt "speed limits"-style framing for regulatory proposals: OpenAI's move — packaging a technical constraint into a memorable political phrase, then running that campaign through the next election cycle — is worth studying as a lobbying technique. But be careful applying it directly to Taiwan: Taiwan has no lame-duck session or midterm election cycle, so the policy-window timing that OpenAI is exploiting doesn't transplant as-is

Cognitive Delta

I used to think the main story in US AI regulation was "Congress is gridlocked, nothing's actually moving." Looking closely at this week's three stories, it turns out the real action was never in Congress at all — a federal agency is raising certainty using laws already on the books, a company is lobbying to claim the vocabulary of future rules before they exist, and a major vendor is using contract leverage to bypass the legislative process altogether. All three are advancing in parallel, which means "no unified federal AI law" has never meant "no one is writing the rules" — the rules are just forming through channels other than legislation, and quickly.

References