Wes Roth

it's all bad now...

⏱ 23 min video · 3 min read30 Jun 2026Worth watching

TL;DR

The US government has banned several frontier AI models (including Anthropic's Claude models and GPT-5.6) while allowing a whitelist of approved companies to continue using them, creating a two-tiered access system. Wes Roth argues this approach is dangerous across multiple dimensions: it entrenches inequality, creates AI safety blind spots, and risks popping the AI investment bubble.

Key points

The US government has banned multiple frontier AI models but granted whitelist access to select companies, creating a two-tiered system where privileged actors gain compounding advantages over everyone else.

Regulating model releases rather than the labs themselves is the wrong target: internal lab research remains unmonitored, meaning a rapid recursive self-improvement event could happen with no public warning.

The delay between internal model capability and public access, previously just months, will now stretch significantly, destroying the public's intuitive sense of AI progress speed.

The investment thesis for AI data centers and compute buildouts assumed global market access; restricting models to US nationals or allies could trigger a repricing of trillions in planned AI infrastructure investment.

Open source AI is not a guaranteed escape route: the government has practical tools (IP blocks, FBI seizures, repository takedowns, GPU-level fingerprinting) to criminalize highly capable open-source models if it chooses to.

Key arguments

→

The Cantillon effect applies to AI: whoever gets early access to frontier models gains compounding advantages in productivity and market position before prices and competition adjust, so monitor whitelist dynamics closely.

→

If you are building an AI investment thesis that assumes a global addressable market, model it again assuming US-only or allied-nation-only deployment, because export control precedents are already set.

→

The correct regulatory analog is a drivers license system: universal access gated by verified identity and demonstrated responsibility, not a class-based system where only elites on the approved list get access.

→

Watch for whether the government course-corrects in coming weeks; a reversal combined with clear, pre-defined criteria for model bans would be a net positive signal for the industry.

→

Projects like Worldcoin's eyeball-scan identity token may become politically viable as governments need a way to verify unique human identity without full ID disclosure to gate frontier model access.

Notable quotes

“We will no longer have an intuitive understanding of AI progress. That's done. It was unlikely that one of these labs would summon a god in one of its data centers without us having some clue. Now, that's on the table.”

“If these models are truly dangerous, then let's not have anyone use them. Not different rules for different tiers of society.”

“It's air conditioning for me, but not for thee. Me and my homies chill up in the penthouse with the air conditioning and super intelligence. That's literally the dystopian scenario that we all warned against.”

Worth watching?

✅

Worth watching the full video?

Worth watching if you want the full emotional force of the argument and Wes's tone, but the key points, risks, and reasoning are all captured here, so skip it if you just need the substance.

Topics

AI & Tech OpenAI

Explore more summaries on these topics →

Saved you some time? The creator still deserves a like.

Watch on YouTube →

More like this