Netflix open-sources video object removal, Anthropic buys a biotech startup, and a data breach hits AI training pipelines

Something quietly shifted in the AI landscape this weekend - and it came from an unexpected direction. While the industry's attention is usually locked on foundation model releases and funding rounds, Netflix's research team just dropped something that Hollywood VFX artists are going to be talking about for a while.

Anthropic's Claude is named after a real person. Which historical figure inspired the name?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

Video editing has a dirty secret: removing an object from footage is trivially easy. Making the scene look like the object was never there - that's the part that takes a VFX team weeks to fix. Pull out a person holding a guitar, and you're left with a floating instrument defying gravity and a background that clearly wasn't reconstructed from scratch.

Netflix's research team built VOID (Video Object Inpainting and Deletion) to solve exactly this. The model doesn't just erase the object - it reconstructs what would physically be behind it, including lighting, shadows, and motion continuity. The result is temporally consistent footage where the object appears to have never existed.

What makes this notable isn't just the capability - it's that Netflix open-sourced it. Any developer, indie filmmaker, or researcher can now access a model that previously would have required significant VFX budget to approximate manually. For content teams working with tight timelines, this could be genuinely transformative.

If you follow the AI video generation space, this is a meaningful step toward production-quality editing workflows that don't require proprietary studio tools.

Read the full story →

While everyone was watching Anthropic's subscriber growth and secondary market activity, the company made a move that signals something bigger about its long-term strategy. Anthropic has acquired Coefficient Bio, a stealth biotech AI startup, in a $400 million all-stock deal - according to reporting from The Information and Eric Newcomer.

Coefficient Bio operated in stealth, so public details are limited. But the acquisition price - $400M in stock - is substantial for a company that hadn't announced itself publicly. It strongly suggests Anthropic sees healthcare and biological research as a core application layer for its models, not just a vertical to market to.

This fits with a broader pattern. The Anthropic team has been vocal about using AI to accelerate scientific research, and acquiring a purpose-built biotech AI team gives them capabilities that are much harder to build from scratch than they are to buy. Expect Claude to get significantly more serious about life sciences applications.

Read the full story →

Meta has paused its work with Mercor, a prominent data vendor used across the AI industry, after a security incident that may have exposed sensitive training data. Multiple major AI labs are now investigating the breach, according to Wired.

Here's what makes this particularly sensitive: the exposed data wasn't just user information. It potentially included details about how AI labs structure their training pipelines - the kind of proprietary information that represents years of research and iteration. That's not just a privacy incident, it's a competitive intelligence risk.

Mercor is a significant player in the AI data supply chain, providing the kind of human-annotated and curated datasets that labs depend on to fine-tune their models. When a vendor at that layer of the stack gets breached, the blast radius extends well beyond Mercor itself.

This is a good reminder that cybersecurity in the AI stack isn't just about model weights and API keys - the data supply chain is just as critical, and often far less scrutinized.

Read the full story →

Utah has approved a one-year pilot that allows an AI system to prescribe and refill psychiatric medications without physician oversight - making it only the second time in US history that clinical prescribing authority has been formally handed to an AI. The Verge reported on this yesterday, and it deserves more attention than it's getting.

State officials frame it as a solution to care shortages and cost barriers - two very real problems in mental health access. The argument is that an AI system can consistently apply clinical guidelines, keep costs down, and reach patients who aren't getting care at all.

Physicians are less convinced. Their concern isn't just efficacy - it's that the system is opaque. Mental health prescribing is highly individualized, and the nuances of why a patient might need a dosage adjustment or a medication switch aren't always captured in structured data. There's also a serious question about whether expanded access will actually reach underserved populations, or just make prescriptions easier for people who already have access.

If you're following the intersection of healthcare AI and AI regulation, this pilot is one to watch closely. The outcome here will influence how much clinical authority other states are willing to delegate to AI systems.

Read the full story →

Anthropic has launched a Political Action Committee, positioning itself to financially back candidates who align with its policy agenda ahead of the midterms. It's a notable escalation in the company's political footprint, and it separates Anthropic further from its earlier image as the safety-first lab that stayed above the fray.

The timing is deliberate. With AI regulation debates heating up in Congress and state legislatures, having a PAC gives Anthropic a direct mechanism to shape which candidates win races where AI policy is on the line. That's different from lobbying - it's direct participation in who holds power.

Combined with the Coefficient Bio acquisition and its strong secondary market position, this is an Anthropic that looks increasingly like a major institutional player rather than a research-forward startup. Whether that's reassuring or concerning probably depends on how much you trust their stated safety commitments to survive the political process intact.

Also worth noting: if you're a developer building on Claude via third-party tools like OpenClaw, Anthropic announced yesterday that Claude subscription limits will no longer cover third-party harnesses as of April 4th - you'll need to switch to pay-as-you-go API access instead. Worth checking your setup if that affects you.

Read the full story →

Quick sidebar: if you're building an AI-powered project and need a website up fast, check out 60sec.site - an AI website builder that gets you from prompt to published in under a minute. Worth bookmarking for your next launch. And for daily AI coverage like this, dailyinference.com has the full archive.

The answer is Claude Shannon! Claude Shannon was the mathematician and electrical engineer who founded information theory - the mathematical framework that underlies all of modern computing and digital communication. Anthropic named their model after him as a nod to the foundational role they hoped the model would play. Shannon published 'A Mathematical Theory of Communication' in 1948, which is arguably the single most important paper in the history of computing.

AI prescribing psychiatric medications without a doctor - does that make you feel hopeful (better access, lower costs) or uncomfortable (opaque system, no human oversight)? There's no wrong answer here. Hit reply and tell me where you land - I read every single response.

That's all for today - see you tomorrow with more. Stay curious out there.