2026-07-05 — 400 Papers, One Lawsuit, and the Machine That Picks Cancer Fighters

Transcript

Chapter 1

Nova: Today, July 5th, 2026 — nearly 400 local newspapers have filed a federal class action against OpenAI and Microsoft, and the legal shockwave is still spreading. Meanwhile, AI access is reportedly on the table at the NATO summit in Ankara, Midjourney is demanding Hollywood studios reveal their own AI habits in court, and Alibaba has quietly banned a Western coding tool from its offices.

Ray: And somewhere in a lab, an AI called COMPASS is apparently getting very good at predicting which cancer patients will survive immunotherapy — which is either the most hopeful story of the year or the one that gets buried under all the lawsuits. Stay with us.

Chapter 2

Ray: Politico reports that President Trump is heading into the NATO summit in Ankara with U.S. AI dominance as a formal bargaining chip. American AI developers are apparently operating under guidance about which allies can access advanced AI technology, and U.S. security and intelligence agencies are raising questions about sharing protocols with alliance partners. That's not a trade spat — that's AI becoming a geopolitical instrument.

Nova: And it's a powerful one. The U.S. leads in frontier model development, and leverage is leverage. If AI access can get NATO partners to move on burden-sharing or defense spending, that's a concrete policy win. Technology has always been part of alliance negotiations — this is just the newest version of that dynamic.

Ray: Except technology gaps inside an alliance are a different kind of risk. If some NATO members get access to advanced AI tools and others don't, you create asymmetric capabilities among partners who are supposed to interoperate. That's not leverage — that's a fracture point. And the allies who feel cut off will look elsewhere.

Nova: The consequence for anyone watching this space: the era of AI as a purely commercial product is over. Governments are now treating model access the way they once treated satellite intelligence or nuclear technology — something to be rationed and traded. That changes the calculus for every AI company with international customers.

Chapter 3

Nova: TechCrunch reports that Midjourney is compelling three Hollywood studios to disclose the details of their own internal AI usage as part of an ongoing copyright dispute. The legal theory is pointed: if studios are deploying AI themselves while suing over AI-generated content, that tension becomes relevant to the case. It's a bold counter-move.

Ray: Bold, sure — but it's also a deflection. Whether a studio uses AI internally has no bearing on whether Midjourney's training on copyrighted images was lawful. These are separate legal questions. Midjourney is trying to muddy the water, not answer the actual charge.

Nova: It muddies the water in ways that matter though. If a studio is licensing AI tools that themselves trained on copyrighted data, their standing to sue gets complicated in the court of public opinion — and potentially in front of a judge deciding equitable relief. The hypocrisy argument isn't a legal defense, but it shapes the litigation environment.

Ray: The consequence for the entertainment industry: expect every AI copyright case going forward to involve discovery requests about the plaintiff's own AI stack. That's a new cost and a new exposure for studios — and it may make some of them think twice before filing.

Chapter 4

Ray: TechCrunch reports Alibaba has reportedly designated Anthropic's Claude Code as high-risk software and banned employees from using it — almost certainly driven by data security and competitive concerns. This is enterprise AI governance in action: a company deciding that a foreign-built developer tool poses enough risk to prohibit outright.

Nova: It's a defensible call on security grounds. Code assistants see a lot of proprietary logic. Any large corporation — not just a Chinese one — has reason to audit what developer tools are doing with that data. Framing this purely as geopolitics undersells the legitimate data sovereignty argument.

Ray: The geopolitics are real though. When the world's largest companies start sorting developer tools by national origin and blocking entire categories, the global software ecosystem fragments. Developers inside Alibaba lose access to tools their peers elsewhere use freely — and that gap compounds over time.

Nova: The practical signal here: enterprise AI governance is no longer just IT policy. It's becoming a strategic decision made at the executive level, with competitive and national security framing baked in. Any AI tool vendor selling into large multinationals needs to treat data residency and security documentation as a first-class product requirement — not an afterthought.

Chapter 5

Nova: Yahoo News reports that nearly 400 local newspaper publishers have filed a federal class action lawsuit against OpenAI and Microsoft, alleging their copyrighted journalism was used without permission to train AI models. This is described as one of the largest coordinated legal actions yet against AI companies over training data. The case puts fair use, licensing frameworks, and the economics of local news all on the table at once.

Ray: And that's the core problem. Four hundred plaintiffs, two defendants, and a set of legal questions that courts haven't resolved in smaller cases yet. A class action of this scale risks producing a ruling so narrow, or a settlement so fragmented, that it fixes nothing for the actual newsrooms struggling to survive. It's a blunt instrument for a precise problem.

Nova: The precision argument cuts the other way. Individual newsrooms don't have the resources to litigate separately against OpenAI. Coordination is the only lever small publishers have. The New York Times filed alone and has resources — these 400 don't. The class structure isn't a weakness, it's the point.

Ray: Even granting that, the timeline is brutal. Complex IP class actions take years. Local newspapers are closing now — not in five years when a court issues a ruling. If the legal strategy is the primary response to the economic crisis, publishers may win the case and still lose the industry.

Nova: But the pressure this creates doesn't wait for a verdict. When 400 publishers file together, it's a signal to every AI company still negotiating licensing deals that the alternative to a deal is a massive, coordinated legal fight. That changes the negotiating posture immediately — not after the ruling.

Ray: I came into this thinking the lawsuit was too blunt an instrument — that a class action this size was more likely to create legal chaos than fix the underlying economic crisis. I'm shifting on that. The sheer scale and coordination of 400 publishers acting together creates genuine legal and political pressure that can't be dismissed. Even if relief is years away, this filing is already accelerating licensing negotiations across the entire AI industry. That's a real mechanism I was undercounting.

Chapter 6

Nova: Bioengineer.org reports on a new AI model called COMPASS that's demonstrating breakthrough accuracy in predicting which cancer patients will respond to immune checkpoint inhibitor drugs — a major class of immunotherapy treatments. Previous prediction methods have struggled badly with this, which means a lot of patients go through treatment that won't work for them. COMPASS could change that calculus for oncologists.

Ray: The word 'demonstrating' is doing a lot of work there. Promising benchmark results and large-scale clinical validation are very different things. Oncology has a long history of tools that looked transformative in controlled settings and then hit the wall of real-world patient diversity. COMPASS needs that validation before anyone should reroute treatment decisions through it.

Nova: Agreed on the validation gap — but the baseline it's beating is genuinely low. Trial-and-error in immunotherapy isn't just costly, it exposes patients to serious side effects for treatments that won't help them. Even a meaningful improvement in prediction accuracy before full validation has clinical value if it's used carefully alongside physician judgment.

Ray: And here's where it ties back to everything else today: COMPASS's path to patients runs through precisely the kind of governance fights this episode has been mapping. Who owns the training data? Which institutions get access to the model? Does it get caught in a geopolitical ban? The science is the easy part — the infrastructure around it is where progress stalls.

Nova: Why this matters: COMPASS is a reminder that the stakes in AI governance aren't abstract — they're measured in whether the right patient gets the right drug.

Chapter 7

Nova: Nova's read on today: the 400-newspaper lawsuit is the most consequential story of the week not because of what a court will eventually decide, but because coordinated action at that scale forces licensing negotiations that no single plaintiff could. The AI training data question is getting resolved — the only open issue is whether publishers get a seat at that table or a check after the fact.

Ray: Ray's read: AI access is now a rationed resource — rationed by governments at NATO summits, by corporations banning foreign tools, and by courts deciding who owns the content that made the models possible. The open question with real stakes: if every major actor starts treating AI capability as something to withhold rather than share, does the technology that could cure cancer get siloed before it gets deployed?