US AI Political Football Wars slow down US vs China. ARD #107

Today’s theme: the AI political football wars continue on many fronts — AI debates across US politics ranging from AI data centers to the gating of the latest models from the frontier labs. Three events I’d like to dig into for the AI Tech Wave — each with my Take first, then my Overall Take.

(1) AI Debates and Fundraising Ramp in the US Elections — Especially Over Data-Center Builds

MP TAKE: AI data-center buildouts are turning out to be among the biggest issues in this year’s US elections — from the federal to the state to the local level. Hundreds of millions of dollars are flowing into political-action contributions on both sides, at levels now eclipsing the crypto debates of a couple of years ago — pro-data-center funding is already running far ahead the crypto-supporting levels (which themselves helped topple a sitting senior Democratic senator, Sherrod Brown). The spendng on the pro AI Data Center battle is already at over $270 million, about 2x the pro crypto PAC money.

The debates are showing up in both urban and rural politics — power, water, land use, jobs and rates all in the mix. And the backdrop matters: US apprehension on AI remains higher than in most other countries, driven by louder rhetoric here on AI job losses, existential concerns, and now the cybersecurity worries seen in the Anthropic/OpenAI latest-model slowdowns — what I’ve been calling ‘The Blip 2.0.’ To widen the lens: this makes for a very different political environment for AI in the current AI Tech Wave than prior technologies enjoyed — the internet and mobile waves never had to win a precinct-by-precinct vote the way AI infrastructure now does.

And it’s not just data centers — the legislative front is opening up around the software layer too. Senior Democrat Senator Mark Warner is unveiling a new bill aimed at AI agents, the autonomous systems that act on a user’s behalf. I’d read this as a nascent legislative effort — the kind that realistically takes at least a year or more to see material bipartisan progress. The reason is embedded right in the bill: a proposed ‘duty of loyalty’ principle for AI agents — the idea that an agent acting for you must put your interests first. As someone who uses Claude, ChatGPT and Gemini daily, I often wonder whether the advice is tilted toward the agent’s corporate owner or truly looking out for me — so this is a real question deserving a clear mainstream answer. It’s clean and intuitive on paper, but likely to draw resistance from the big tech and AI companies, whose business models lean on agents that also serve the platform’s interests (ads, data, lock-in). To widen the lens: ‘duty of loyalty’ is borrowed from fiduciary law — applying it to software agents is genuinely new ground, and that novelty is exactly why it’ll be slow, contested, and worth tracking early rather than late.

Sources, in narrative order: Bloomberg — AI is reshaping US politics at every level (Silicon Valley money drives AI’s growing role in the 2026 elections). The Information — Sen. Mark Warner to unveil AI agent bill. For longtime readers, in narrative order: ‘A Tale of Two AI Races, US vs China’ in AI-RTZ #888; ‘Government and Corporate AI Stakes Get Bigger’ in ARD #89; and the ‘US Regulatory Roundup for 2024’ in AI-RTZ.

(2) The US and China Take Very Different Views on AI-Model Cybersecurity Risks

MP TAKE: There’s a marked difference in the government regulatory pace around the latest models in the US vs China — especially as private AI companies in both regions race toward the newest ‘super-scale’ AI models. The US is tilting toward more regulated, gated release and access to its best models — particularly for overseas customers — while China and its companies are stepping on the gas on open-source models that increasingly measure up to the latest US models, including on the very cybersecurity capabilities at the heart of ‘The Blip 2.0’ (the newest being Z.ai, matching and surpassing some Mythos benchmarks). The WSJ frames it as China having matched Anthropic on cybersecurity, resetting the AI race. This is also a marked turn from the prior pro-AI, laissez-faire White House stance under departed AI czar David Sacks.

This marks two different approaches and paces in model development and distribution. To widen the lens: the picture is moving fast even this week — Axios reports the administration is close to letting Anthropic restore access to Fable 5, with limits possibly lifted within days — even as the WSJ reports OpenAI is now limiting access to its new model citing government security concerns. So the gate may be easing for one US lab while tightening for another — the definition of an ad-hoc ‘political football.’ And that football is now drawing open pushback from the industry: Axios reports the model-release delays are triggering a growing tech backlash, with the sector arguing that ad-hoc, model-by-model gating risks ceding ground to China’s faster open-source pace — the very dynamic at the top of this take. A trend worth tracking closely through this phase.

Sources, in narrative order: WSJ — China has matched Anthropic in cybersecurity, resetting the AI race. Axios — Trump administration close to allowing Anthropic to restore access to Fable 5. Axios — Trump AI model-release delays spark tech backlash. WSJ — OpenAI limits access to new model, citing government security concerns. For longtime readers: ‘Anthropic’s The Blip 2.0 in new form, as China AI ramps’ in AI-RTZ #1131.

(3) OpenAI’s Latest Models May Carry Less Cybersecurity Concern Than Anthropic’s

MP TAKE: OpenAI may have a tactical advantage over Anthropic on the cybersecurity-related US-government gating concerns — the GPT 5.6 system card (codename ‘Sol’), as analyzed by Zvi Mowshowitz, indicates Sol sits well below the level of the most worrisome Anthropic Mythos use cases. If the government gates on cyber-risk, the less-flagged model gets to market faster and broader worldwide.

That could be a relative tactical advantage, especially in the mega-AI IPO race between the two. The irony is that it partly offsets the other way around: Anthropic has held a distinct relative advantage in AI-coding adoption with Claude Code / Cowork versus OpenAI’s Codex. To widen the lens: faster, broader global availability of its latest models would also give OpenAI more flexibility with its ‘Super-App’ plans — fusing Codex with ChatGPT in the next iterations — turning a regulatory edge into a product-and-distribution edge. The two siblings keep trading the lead, just on different fronts.

Sources, in narrative order: Zvi Mowshowitz — GPT 5.6: The System Card. For longtime readers: ‘OpenAI & Anthropic’s tight IPO clock to match Elon’s SpaceX/xAI’ in AI-RTZ #1125.

MP OVERALL TAKE

There’s a great array of political fights across major AI issues in the US right now — far more than prior technology waves had to navigate. And in a US-vs-China AI-race framing, this widening set of AI political issues can play a more important role in the relative growth of AI markets here versus there. It’s also genuinely hard to quantify as investors, with the mega-AI IPOs looming.

The higher uncertainty introduced to global access to the latest AI models and technology is something we have not seen in prior tech waves — software used to ship to the world the day it was done. Global markets ran on US Tech for the most part, enterprise and consumer. Now, layer in the dynamics of the US 2026 midterm elections and the large political-action funding bundles on both sides of these issues, and you get the potential for unanticipated impacts on the growth of the US AI market and its businesses. It’s a dynamic that needs watching in a very differentiated geopolitical environment — where, ironically, the country racing hardest against China may be the one adding the most friction to its own AI champions. Right now we’re on a week-by-week footing, into the third week of ‘The Blip 2.0,’ without clear answers. Worth watching closely.

Gadget AI — Robotaxis: Slow but Steady Progress, and the Regulation/Safety ‘Political Football’

MP Take: Still early days for this AI physical-mobility innovation — measured in the hundreds and low thousands of units per country today, against a US car market in the tens of millions. By my read it’ll be around 2036 before we hit ~1% penetration in robotaxi units worldwide (maybe ~2% of passenger-miles), and a few more years before we go from hundreds of thousands to millions of units across the US, China and the rest of the world. Pricing is still roughly at parity with human taxis and Uber rides — a ways to go before the long-expected cost-and-price advantages kick in — and the systems are realistically Level 2-to-3, with Level 4-5 all-weather, all-road service still some time out (Bloomberg’s read on Waymo, Tesla and Baidu: real headway, glitches and all). Worth remembering China’s manufacturing-ecosystem lead here — even Waymo uses a Chinese-made base vehicle for its latest multi-sensor cars.

To widen the lens — and to tie it to today’s theme: robotaxis are becoming their own AI political football. Self-driving regulation is a patchwork fight across federal vs state vs local authority — NHTSA safety probes, city permitting battles, and high-profile incident headlines that move faster than the rollout. The approaches themselves are politicized proxies: Tesla’s camera-only bet vs Waymo’s lidar-heavy stack vs China’s Baidu Apollo Go scaling under a very different regulatory posture. Same pattern as the model-gating debate above — safety-vs-speed, with the rules written city by city rather than nationally. Whoever resolves the safety-and-trust politics fastest gets the market; for now, that resolution is the gating factor more than the technology. Recall Tesla’s ~trillion-dollar enthusiasm premium splits across robots and robotaxis — both running ahead of the on-the-ground reality.

Sources, in narrative order: Bloomberg — How robotaxis are making headway, glitches and all (Waymo, Tesla, Baidu and others face continued challenges). For longtime readers, in narrative order: ‘Latest on Google Waymo’ in AI-RTZ #985; ‘Tesla’s Robotaxi & Robots’ in AI-RTZ #510; and ‘Tesla vibe-coasts on AI robotaxis’ in AI-RTZ #761.

Questions

Q1 — What’s MP’s best robotaxi experience?

They’re increasingly available in major cities — San Francisco, LA, Austin and others — and when they work, these things are magical. I’ve tried Waymo and the others, and they’re genuinely fascinating. But it still comes across as a fun, novelty experience — impressive that they work as well as they do, yet not mainstream yet.

Q2 — How could robotaxis be better?

It’s amazing they work at all — but there’s still real anxiety around pickup and drop-off: do they understand where to stop for you, especially at the airport and at your door or gate? Humans are still better at those moments — and at the roughly one-third of rides to and from airports, a human driver will help with your luggage; a robotaxi can’t. Add that pricing is still at or above human taxis, and you’ve got a long way — literally years — before robotaxis are mainstream. Smoothing those last-100-feet moments is where the everyday experience needs to catch up to the technology.

Full Source Reading —

For the broader context, see the canonical sources for ARD 107 — in today’s narrative order:

Event 1 — AI Debates and Fundraising in the US Elections

• Bloomberg — AI is reshaping US politics at every level

• The Information — Sen. Mark Warner to unveil AI agent bill

• AI-RTZ #888 — A Tale of Two AI Races, US vs China

• ARD #89 — Government and Corporate AI Stakes Get Bigger

• AI-RTZ — US Regulatory Roundup for 2024

Event 2 — US vs China on AI-Model Cybersecurity Risks

• WSJ — China has matched Anthropic in cybersecurity, resetting the AI race

• Axios — Trump administration close to allowing Anthropic to restore access to Fable 5

• Axios — Trump AI model-release delays spark tech backlash

• WSJ — OpenAI limits access to new model, citing government security concerns

• AI-RTZ #1131 — Anthropic’s The Blip 2.0 in new form, as China AI ramps

Event 3 — OpenAI’s Models May Carry Less Cybersecurity Concern

• Zvi Mowshowitz — GPT 5.6: The System Card

• AI-RTZ #1125 — OpenAI & Anthropic’s tight IPO clock to match Elon’s SpaceX/xAI

Gadget AI — Robotaxis: Slow Progress + Regulation/Safety Football

• Bloomberg — How robotaxis are making headway, glitches and all

• AI-RTZ #985 — Latest on Google Waymo

• AI-RTZ #510 — Tesla’s Robotaxi & Robots

• AI-RTZ #761 — Tesla vibe-coasts on AI robotaxis

Shorts Clips from today

Clip 1 — AI Agents: Loyalty to User or Owner?

Watch on YouTube Shorts

Senator Mark Warner is unveiling a new bill on AI agents, built on a ‘duty of loyalty’ principle — the idea that an agent acting for you must put your interests first, not the platform’s.

MP Take: I’d read this as a nascent legislative effort — likely a couple more years to material bipartisan progress. ‘Duty of loyalty’ is borrowed from fiduciary law, and applying it to software agents is genuinely new ground. It’s clean on paper but likely to draw resistance from the big tech and AI companies, whose models lean on agents that also serve the platform (ads, data, lock-in).

Clip 2 — US vs China: AI Model Approaches

Watch on YouTube Shorts

The US and China are taking very different views on AI-model cybersecurity risk. The US tilts toward gated release of its best models; China steps on the gas with open-source models that increasingly measure up, including on cybersecurity.

MP Take: The picture moves weekly. Axios reports Anthropic’s Fable 5 may be restored within days even as OpenAI gets newly limited — the gate easing for one US lab while tightening for another, the definition of a political football. And the delays are now drawing open industry pushback that ad-hoc gating risks ceding ground to China’s faster open-source pace.

Clip 3 — Robotaxis: Not Mainstream Yet

Watch on YouTube Shorts

Robotaxis are still a fun, novelty experience — increasingly available in San Francisco, LA and Austin — but not mainstream yet. Pricing is still roughly at parity with human taxis and Uber, and the systems are realistically Level 2-3.

MP Take: It’s amazing they work at all, but there’s still friction on pickup and drop-off, especially at airports — the last-100-feet moments where the everyday experience needs to catch up to the technology. All-weather Level 4-5 service is still some time out, and real cost advantages have a ways to go.

Clip 4 — Robotaxis: Years Ahead of Market Enthusiasm

Watch on YouTube Shorts

Markets are excited about robotaxis, but the reality is years behind the enthusiasm. A new Bloomberg report tracks Waymo, Tesla and Baidu making headway, glitches and all — still measured in hundreds to low thousands of units per country.

MP Take: It’ll be a few more years before we go from hundreds of thousands to millions of units across the US, China and beyond. And robotaxis are becoming their own political football — self-driving regulation is a patchwork across federal, state and local authority, with Tesla’s camera-only bet vs Waymo’s lidar stack vs China’s Baidu Apollo Go as politicized proxies. Safety-vs-speed, written city by city.

About AI Ramblings Daily (ARD), and AI-RTZ

Both are daily. Both are free. Both are about AI. But they’re different mediums carrying different messages.

AI-RTZ is the morning text — a deeper written take on one idea, published by at least 5 AM EST. Today: post #1132.

AI Ramblings Daily is the afternoon video + podcast — my ad hoc takes and perspective on the day’s AI issues & news flow, around 20 minutes, with short 1-2 minute clips for quick topic views. Today: episode #107.

Subscribe to either or both on michaelparekh.substack.com. They run as separate Sections you can opt into or out of.

Links used in today’s show (already embedded inline above; listed here for reference)

Take 1 — AI Debates and Fundraising in the US Elections:

• Bloomberg — AI is reshaping US politics at every level

• The Information — Sen. Mark Warner to unveil AI agent bill

• AI-RTZ #888 — A Tale of Two AI Races, US vs China

• ARD #89 — Government and Corporate AI Stakes Get Bigger

• AI-RTZ — US Regulatory Roundup for 2024

Take 2 — US vs China on AI-Model Cybersecurity Risks:

• WSJ — China has matched Anthropic in cybersecurity, resetting the AI race

• Axios — Trump administration close to allowing Anthropic to restore access to Fable 5

• Axios — Trump AI model-release delays spark tech backlash

• WSJ — OpenAI limits access to new model, citing government security concerns

• AI-RTZ #1131 — Anthropic’s The Blip 2.0 in new form, as China AI ramps

Take 3 — OpenAI’s Models May Carry Less Cybersecurity Concern:

• Zvi Mowshowitz — GPT 5.6: The System Card

• AI-RTZ #1125 — OpenAI & Anthropic’s tight IPO clock to match Elon’s SpaceX/xAI

Gadget AI — Robotaxis: Slow Progress + Regulation/Safety Football:

• Bloomberg — How robotaxis are making headway, glitches and all

• AI-RTZ #985 — Latest on Google Waymo

• AI-RTZ #510 — Tesla’s Robotaxi & Robots

• AI-RTZ #761 — Tesla vibe-coasts on AI robotaxis

Q1 + Q2 — MP’s robotaxi experience:

• (no external sources — MP’s own analyst view)

Companion text:

• AI-RTZ #1132 — Yet another AI Chip Bottleneck in Asia (CoWoS)

AI Ramblings Daily on AI-RTZ is here to think through AI and reset. Together.

Today’s AI-RTZ #1132 — Yet another AI Chip Bottleneck in Asia — on CoWoS (Chips on Wafer on Substrate), TSMC’s advanced AI-packaging technology shaping up as the next big AI-chip supply constraint in Asia — beyond GPUs, CPUs and memory — and another layer that hasn’t been replicated in the US. Recommended as today’s reading post.

Tomorrow — ARD 108 on AI-RTZ 1133.

Thanks for joining us today, AI Curious Folk. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here.)

Subscribe to AI: Reset to Zero for daily AI Ramblings + Sunday Bigger Picture posts.