AI companies have been arguing that training on copyrighted material is fair use. That argument is now being tested in federal court, with a named CEO, a coalition of major publishers, and a complaint that quotes the company's own internal motto back at it. The same week, a Microsoft feature designed to give your computer a photographic memory turned out to store everything it photographed in a plaintext database. And GameStop announced a $55 billion acquisition of eBay with roughly a quarter of the capital required to close it. The thread running through all three is the same: institutions built to slow things down are finally catching up to companies that bet they wouldn't.
Meta's Llama Training Was Built on Books It Didn't Pay For. Now the Publishers Are Suing.
What Happened
Five major publishers (Elsevier, Cengage, Hachette, Macmillan, and McGraw Hill) filed a federal lawsuit in Manhattan accusing Meta of using millions of copyrighted books and journal articles without permission to train its Llama language model. The complaint names CEO Mark Zuckerberg personally, alleging he authorized and encouraged the infringement. Meta says it will fight the case, arguing AI training on copyrighted material qualifies as fair use.
Why It Matters
Developers building with or on top of LLMs should watch this case: a ruling against Meta would force the industry to rethink how training data is sourced and licensed, with direct implications for every model in production. Anthropic settled a similar suit for $1.5 billion last year, which gives the publishers a credible roadmap.
Source: Variety
Tags: AI Tooling, Industry, Engineering Practice
Microsoft's Copilot+ Recall Stores Your Passwords in Plaintext. The Company Says It's Fine.
What Happened
Security researcher Kevin Beaumont demonstrated that Microsoft's Copilot+ Recall feature stores OCR-processed text from continuous screen captures in an unencrypted SQLite database in the user's local folder. Other accounts on the same device can access it, and malware can exfiltrate months of browsing history and passwords in seconds. Microsoft has acknowledged the finding and maintained it is not a security concern.
Why It Matters
For security engineers, this is a textbook third-party risk and data centralization case study: the vulnerability requires no exploit, just access to a predictable local path. Recall ships enabled by default on new Copilot+ devices, meaning millions of users are exposed unless they opt out during setup.
Source: Windows Central
Tags: Security, Dev Tools, Industry
GameStop Bid $55.5 Billion for eBay. Investors Weren't Convinced.
What Happened
GameStop made an unsolicited offer to acquire eBay for approximately $55.5 billion ($125 per share, split evenly between cash and stock), representing a 20 percent premium to eBay's Friday close. CEO Ryan Cohen cited a $20 billion financing letter from TD Bank and GameStop's $9.4 billion cash pile, leaving a substantial gap unaccounted for. GameStop's stock dropped 10 percent on the news; eBay shares rose to around $109, well below the offer price.
Why It Matters
The bid's collectibles angle is worth tracking for developers and creators who sell gaming merchandise, limited editions, or fan-created content on eBay: a GameStop acquisition could redirect the platform's current focus on trading cards, collectibles, and luxury goods in ways that affect seller communities and category tooling.
Source: CNBC
Tags: Industry, Career
The Bigger Picture
The common read on weeks like this one is that tech is finally being held accountable. That framing is tempting, but it is probably too clean. Meta will fight the copyright case on fair use grounds, and the legal outcome is genuinely uncertain. Microsoft's response to Beaumont's findings suggests the company does not consider plaintext local storage a security failure under its current threat model, which is its own kind of policy position. GameStop's bid may be a negotiating opening, a PR move, or a real attempt at an acquisition by a CEO who has surprised skeptics before.
Each story has a version where the institution catches up and a version where it doesn't. What is actually happening is that the gap between what these companies built and what the rules were designed to handle is becoming visible in ways that are harder to manage quietly. That gap has been there for years. Courts, regulators, and researchers are just getting better at finding the edges of it. For developers working in AI tooling, security, or platform-dependent product categories, the practical question is not who wins these specific cases. It is how the incentives shift as the cases accumulate.
For a closer look at the security principles the Recall case puts on display, cybersecurity certifications that cover threat modeling and data exposure are worth reviewing. And if the AI training data question has you thinking about which tools are actually worth building on right now, the best AI tools ranked by practical use covers the landscape as it stands.
