Daily Briefing — May 18, 2026

Tech companies are gutting their middle management layers and pointing at AI as the reason. Coinbase just cut 14% of its workforce with the explicit goal of flattening its structure. They are not alone. Amazon, Meta, and Block have all done versions of this over the past year, removing management tiers at scale and framing it as operational efficiency unlocked by AI tools.

The framing leaves out what middle managers actually do. They sponsor your promotion, advocate for your work in rooms you are not in, and absorb the organisational chaos so you can focus on shipping. When that layer disappears, the work does not vanish with it — it either falls on you, or it just stops happening.

Researchers tracking this shift are flagging the same thing: AI is moving tasks that used to sit with managers directly onto individual contributors. Wider spans of control, more self direction, less institutional support. That is a real problem for early career workers in particular. The informal mentorship and career scaffolding middle managers provided is being stripped out, and nobody is putting anything in its place.

Quick recap of what Anthropic is trying to do here. Last year a group of authors sued the company over the well-documented fact that it trained Claude on a corpus of pirated books pulled from shadow libraries like LibGen. Rather than drag the case through years of trial, Anthropic offered to settle: $1.5 billion paid into a class fund, distributed to roughly 500,000 authors whose work showed up in the training data. The pitch was that this would be the cleanest possible exit — pay a record-setting number, close the chapter, set a template the rest of the AI industry could follow when their own training-data lawsuits inevitably land.

It was supposed to be a landmark moment. The largest copyright settlement in US history, wrapped up neatly, everyone goes home. Except a federal judge is now pumping the brakes. US District Judge Araceli Martinez-Olguin declined to approve the deal after authors and class members raised pointed questions about where the money was actually going.

Look at the math. The lawyers are asking for over $320 million in fees. Each author in the class gets about $3,000. You do not need a law degree to see why people are upset. One objector put it plainly: every dollar counsel takes is one not going to the people who were actually harmed.

It gets messier. The authors' legal team has been accused of actively trying to shut objectors out from voicing concerns in court. A judge noticing that and asking for an explanation is meaningful. Even in the biggest AI copyright case so far, who gets paid and how much is still being fought over in plain sight.

This case is shaping how creators, platforms, and AI companies will negotiate the value of human work going forward. The outcome here sets a precedent that will echo well beyond publishing.

Instructure, the company behind the Canvas learning platform used by roughly 275 million students and staff across 9,000 schools worldwide, got hit by a ransomware attack from a group called ShinyHunters. After a week of outages, defaced login pages, and delayed assignment deadlines, Instructure announced it had "reached an agreement with the unauthorised actor." Nobody is saying ransom out loud, but everyone in the industry is reading between those lines just fine.

The data at stake was not trivial. We are talking 3.6 terabytes of student IDs, email addresses, names, and messages. That is the kind of haul that shows up in credential stuffing attacks and phishing campaigns for years after the original breach. The people most at risk are not executives with breach response teams on speed dial. They are students and school staff who had zero say in how their data was stored or protected.

Governments universally advise against paying ransom. Companies pay anyway, all the time. Payment does not guarantee the data gets deleted, and it does guarantee the attackers learn this particular target will negotiate next time. The "agreement" language Instructure used is the corporate equivalent of not making eye contact while sliding an envelope across a table.