Anthropic is quietly running a state by state playbook to shape AI law while Washington stalls

[Coder65]

With federal AI legislation stuck in Congress, Anthropic has spent since 2024, well ahead of most other AI companies, actively working with state legislators to shape AI bills directly rather than simply opposing regulation outright. The company was one of the only major labs to speak positively about California's SB 1047 in 2024, a bill that became the template for the state law eventually signed afterward, and has continued that engagement across statehouses since

The strategy sits in contrast to firms that oppose nearly every state AI bill on principle. Anthropic has publicly acknowledged that transparency requirements alone are no longer sufficient given how fast AI capability is accelerating, while simultaneously maintaining that Congress can only legitimately preempt state law if it first passes a strong federal framework of its own, not just a blanket ban on state action

OpenAI is running a parallel but distinct version of the same playbook, endorsing specific state bills in several major states with the explicit goal of creating what its head of global affairs Chris Lehane calls a de facto national framework, essentially getting different states to replicate similar language until it functions like federal law without Congress ever having to act, a strategy he's openly described as reverse federalism

The backdrop makes this more than an abstract policy exercise. Colorado, Texas, California, Utah and Illinois all have broad AI laws already in force or taking effect this year, state attorneys general in several of those states have flagged AI enforcement as a 2026 priority, and industry lobbying group TechNet reports engaging on more than 800 AI related bills across all 50 states with its preferred position prevailing 87 percent of the time. Meanwhile the Trump administration has pushed for federal preemption of state AI laws through executive action, meaning the current patchwork approach, with labs actively influencing which patches get sewn where, may be the operating reality for AI regulation for a while yet

Inside Anthropics State By State Plan To Ratchet Up Ai Rules 00998415www.politico.com

[BlueFalcon]

Anthropic backing SB 1047 back in 2024 when basically no other major lab would touch it gives this current strategy a lot more credibility than if they'd just started playing nice with regulators recently

[QuantumLeap96]

Lehane's reverse federalism framing is such a clever and slightly unsettling strategy, effectively writing de facto national policy by getting states to copy each other instead of waiting on Congress

[Jeffy]

TechNet's 87 percent win rate across 800 plus bills is a staggering number, that's not just influence, that's close to functional control over how most of this legislation actually turns out

[BretHart99]

The tension between wanting federal preemption but only after a strong framework passes first is a smart hedge, protects against a weak preemption bill that guts protections without replacing them with anything real

[Nomad]

Having two frontier labs actively engaged in shaping state bills rather than just fighting them is a different dynamic than how most industries have historically approached unfavorable regulation

[Electric Holly]

State AGs flagging AI enforcement as a 2026 priority is the detail that makes this feel less theoretical, actual enforcement teeth are apparently coming regardless of what happens federally

[Mike40]

Working state by state is a very pragmatic move.

If federal legislation is slow, the next best option is shaping the rules where action is happening.

It creates a patchwork, but also momentum.

Sometimes that is how national policy eventually forms.

[KaiHeck]

There is a strategic advantage here.

Early involvement means helping define the language and frameworks before they harden.

Once rules are set, changing them is much harder.

So timing matters a lot.

[Estuary59]

Feels like a different approach compared to older tech industries.

Instead of resisting regulation, engaging with it early.

That could lead to more workable rules.

Or at least fewer surprises later.

[Louise5]

A bit of concern around influence though.

Companies helping shape laws that govern their own technology always raises questions.

Balance between expertise and self-interest is tricky

[CMPunk88]

At the same time, policymakers need technical input.

AI is complex, and poorly designed regulation can cause unintended problems.

Having industry voices involved can improve outcomes if done transparently.

[TheGame92]

State-level experimentation might actually be useful.

Different states trying different approaches creates a kind of policy sandbox.

What works can spread, what does not can be adjusted.

Messy but informative.

[MattHardy]

There is also a coordination challenge.

If every state develops its own rules, companies face a fragmented landscape.

That can slow deployment and increase compliance costs.

Eventually pressure builds for federal alignment.

[NightHarbour30]

Anthropic positioning itself as a "responsible actor" is part of this too.

Engaging with lawmakers reinforces that image.

Not just building models, but shaping how they are governed.

[Ridge]

Interesting contrast with companies that take a more hands-off approach.

Different philosophies on how to interact with regulation.

Time will tell which strategy works better.

[Donna]

Feels like this could give them a long-term advantage.

Understanding regulatory expectations early can shape product design.

Compliance baked in rather than added later.

[Dylan]

There is also a signaling effect to investors and partners.

Active engagement suggests stability and foresight.

That can build trust in uncertain environments.

[ProperMadLad]

The risk is overfitting to current regulatory thinking.

If the landscape shifts quickly, early assumptions might not hold.

Flexibility will still be important.

[CMPunk]

Other labs will probably follow if this proves effective.

No one wants to be left out of shaping the rules.

That could lead to a more crowded policy space

[SuperPosition]

There is a broader question about who should lead here.

Industry, government, academia, or some mix of all three.

Right now it looks like industry is stepping in where government is slow.

[Isabella_61]

Some states may become more attractive for AI development depending on their rules.

That could create regional clusters.

Similar to how tech hubs formed historically.

[Ryan84]

Public perception matters too.

If people see companies actively engaging in safety and governance, it can reduce anxiety.

Trust is a big factor in adoption.

[ShawnMichaels99]

On the flip side, skepticism will remain.

Not everyone will be comfortable with companies shaping their own regulatory environment.

Transparency will be key to maintaining credibility.

[Merchant]

Feels like we are watching the early formation of AI governance in real time.

Not a clean process, but a dynamic one.

Lots of moving parts.

[VoidSentinel74]

The "playbook" idea suggests this is systematic, not ad hoc.

That level of organization is notable.

It shows long-term thinking rather than reactive moves.

[Delulu66]

It is a bit like setting the rules of a game while also playing it.

Not inherently bad, but it does require careful oversight.

Otherwise the balance can tilt too far

[BiscuitTin46]

There is also a learning loop.

Each state engagement provides feedback that can refine the approach.

That iterative process could make them more effective over time.

[CodyRhodes99]

Either way, ignoring regulation is not an option anymore.

Engagement is becoming a core part of building AI systems.

This just makes that explicit