2026-06-18 — Grok at War, Anthropic Shut Out, and Who Owns the Machine

Transcript

Chapter 1

Nova: Today, June 18th, 2026 — sworn court testimony alleges the Pentagon used Elon Musk's Grok AI to support the targeting of thousands of munitions against Iran in under four days. Meanwhile, the White House handed Anthropic an ultimatum security experts call technically impossible, and Bernie Sanders wants every American citizen to own a slice of the AI industry. Midjourney is building an ultrasound scanner, and a new Pew study reveals that half of Americans use chatbots but barely one in six thinks AI is good for society. The thread running through all of it: who actually controls this technology — and what happens when the answer is no one?

Chapter 2

Nova: The Verge reports that the Trump administration abruptly ordered Anthropic to cut off all foreign nationals from its Fable 5 and Mythos 5 models — and that includes foreign nationals working inside the US and even Anthropic's own employees. The trigger, reportedly, was concern over SK Telecom's alleged ties to China. Then the White House added a condition for restoring access: Anthropic must prove it can block every jailbreak. Security experts say that is not a bar anyone can clear. It is, effectively, a permanent shutdown dressed up as a security requirement.

Ray: The government has a legitimate interest in keeping frontier AI out of adversarial hands — that part is not unreasonable. Nation-state actors probing these models for dual-use capabilities is a real threat. The problem is the specific demand. Asking for a provably jailbreak-proof model is like demanding a lock that cannot be picked before you're allowed to sell locks. It sets an impossible technical standard that conveniently achieves the same result as an outright ban while giving the administration political cover.

Nova: And the fallout extends well beyond Anthropic's business. Macron and Modi have both raised alarms publicly — the US just demonstrated it can sever access to American AI infrastructure overnight, for any partner, for any reason. That is the concrete consequence for every government, every enterprise, every researcher outside US borders relying on American frontier models. The AI sovereignty argument that used to sound theoretical now has a very specific case study attached to it.

Ray: That concern is real, but it cuts both ways. If the alternative is leaving models with alleged national security vulnerabilities freely accessible, allies alarmed about US control should perhaps be more alarmed about what adversaries could do with unrestricted access. The export control instinct is not wrong — the execution here is what's indefensible. Those are different problems, and conflating them lets the administration off the hook for demanding the impossible.

Chapter 3

Nova: The Washington Post reports that Senator Bernie Sanders has unveiled a legislative plan to give American citizens direct ownership stakes in the country's largest AI companies. The framing is explicit: AI is generating enormous wealth, and Sanders argues that wealth should be democratized rather than concentrated in a handful of firms and investors. The Post describes it as the most ambitious left-wing policy intervention into AI governance yet introduced in Congress.

Ray: Government ownership stakes in private AI labs is a structural change with serious second-order effects. Research priorities at frontier labs are already shaped by investor pressure — layer in political accountability and you introduce a different kind of distortion. Elected officials have incentives that have nothing to do with long-term safety research or capability development. The investment climate for frontier AI, which requires enormous capital commitments on uncertain timelines, could chill fast.

Nova: TechCrunch covered the Pew Research data that gives Sanders real political wind here. Forty-nine percent of Americans now use chatbots at least occasionally — up from thirty-three percent in 2024 — but only sixteen percent believe AI will have a positive societal impact, and sixty-three percent say the technology is advancing too fast. That is not a niche concern. That is a majority of the country using a technology they fundamentally distrust. That gap does not stay a gap — it becomes policy pressure.

Ray: People use technologies they distrust all the time. Social media, credit scoring, surveillance cameras — adoption and approval have never been tightly coupled. The Pew numbers are striking, but the historical pattern suggests the gap can narrow as tangible benefits accumulate rather than through structural ownership changes. Sanders' plan might be a political signal more than a workable mechanism.

Nova: The difference is that those other technologies did not generate wealth at this concentration or speed. When nearly half the country is actively using a product but fewer than one in six think it benefits society, that is a specific kind of political combustion. Whether Sanders' bill passes or not, it is defining the terms of the next intervention — and something with real teeth will follow if the trust numbers do not move.

Chapter 4

Nova: Ynetnews is reporting on sworn court testimony alleging that the Trump administration deployed xAI's government-version Grok model to support the targeting and firing of thousands of munitions against Iran — all within a 96-hour window. This comes from a court filing, not a leak or anonymous source. If the testimony holds, it represents one of the most significant disclosures of a commercial AI system used in alleged active combat operations ever made public.

Ray: Sworn testimony is not the same as verified fact. The filing alleges this — it does not establish the full scope of what Grok actually did versus what human operators decided. The critical question that testimony alone cannot answer is where the AI's role ended and where human judgment began. 'Support' is doing a lot of work in that sentence. Did Grok flag targets? Generate recommendations? Automate firing sequences? Those are categorically different things, and collapsing them into one alarming headline obscures the oversight question entirely.

Nova: That ambiguity is precisely the problem. Congress is being asked to provide oversight of military AI deployment without knowing what 'support' means in operational terms. The testimony has fueled urgent concern on the Hill because the disclosure itself reveals there is no public framework, no announced doctrine, no congressional authorization for deploying a commercial AI product in this context. The oversight gap is not theoretical anymore. It showed up in a court filing.

Ray: The oversight gap is real. But using an unverified account to push through sweeping new legislative mandates on military AI carries its own risks — you could lock in frameworks built on a misunderstanding of what actually happened. The right response is a full congressional inquiry before legislation, not legislation triggered by sworn-but-unconfirmed testimony.

Nova: Here is the problem with waiting for full verification: the pace alleged in this testimony — thousands of munitions, 96 hours — means any future deployment could be over before an inquiry concludes. If the AI is operating at that speed, the oversight has to be built in before the operation, not investigated after. That is a structural argument for guardrails that does not depend on whether every detail in this specific filing is confirmed.

Ray: I came into this chapter insisting the sworn testimony was unverified and that the human-in-the-loop question remained unresolved — skeptical that this disclosure alone justified sweeping new oversight mandates on military AI. I held that position until this structural point. If the testimony is even partially accurate, the speed and scale of alleged AI-assisted munitions targeting at that pace makes immediate legislative guardrails on military AI deployment a necessity I can no longer argue against. Guardrails have to precede deployment. I concede that.

Chapter 5

Ray: The Verge reports that Midjourney CEO David Holz has unveiled the company's first hardware product — the Midjourney Scanner, an ultrasound-based full-body imaging device — alongside plans to open a spa in San Francisco. This is an AI image generation company announcing it will scan human bodies for medical and wellness purposes. The regulatory pathway for a medical imaging device involves FDA clearance, clinical validation, and demonstrated safety data. Midjourney has none of that history.

Nova: The pivot is audacious, but Midjourney is not the first software company to decide that hardware and health are the next frontier. The argument for letting AI-native companies enter this space is that they bring different capabilities — particularly in image interpretation — that incumbents lack. Regulatory pathways are messy, but they exist precisely for new entrants. The question is whether Holz is serious about navigating them or treating FDA clearance as a detail to sort out after the spa opens.

Ray: That last part is the concern. 'Wellness hardware' is a category that has historically been used to sidestep medical device regulation — market it as a spa experience, avoid clinical claims, and operate in the gap. If Midjourney's scanner is making any kind of health-relevant imaging available to consumers without clinical validation, the harm potential is real. Misread ultrasounds, false reassurance, delayed diagnosis — those are not hypothetical risks in an unvalidated imaging product.

Nova: And here is where this story connects to everything else in today's episode. The same accountability vacuum that allowed Grok to allegedly support munitions targeting without a public framework, the same overnight cutoff power the White House exercised over Anthropic — that pattern applies directly here. When any AI company can unilaterally enter the medical device space without a clear regulatory trigger, the governance question is identical. The technology moves; the oversight structures do not. That is the through-line.

Chapter 6

Nova: Today's stories share a single fault line: capability is outrunning every institution designed to govern it — export controls, military oversight, financial regulation, medical device law. The countries and companies that build durable governance frameworks now will set the terms for everyone else. The ones that don't will keep finding out what went wrong from court filings.

Ray: The Grok testimony changed something for this analysis today. The open question with real stakes is this: will Congress pass binding pre-deployment authorization requirements for military AI before the next alleged operation is disclosed — or will the oversight framework always arrive one court filing too late?