Google DeepMind workers unionize, AI hiring algorithms under fire, and Cerebras eyes $26B IPO

Two stories landed yesterday that, at first glance, seem completely unrelated. One is about AI workers pushing back on how their tools get used. The other is about an AI tool publicly destroying someone's reputation. But they're telling the same story - when AI systems operate without proper oversight, real people pay the price.

Let's get into it.

Cerebras Systems - the AI chip company now heading for a blockbuster IPO - is known for its unusual chip design. What makes the Cerebras WSE (Wafer Scale Engine) so different from standard AI chips?

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

In a significant escalation of employee activism inside one of the world's most powerful AI labs, Google DeepMind workers in the UK have voted to unionize - and the timing is directly tied to the company's military AI partnerships.

Workers cited Google's recent deal with the US military as a core concern. One employee pointed to the Pentagon's ongoing feud with Anthropic and the situation in Iran as evidence that the Department of Defense is "not a responsible partner." The letter from workers was slated to go to management on Tuesday.

This isn't the first time AI lab employees have pushed back on military contracts - Google's Project Maven protests in 2018 are the most famous example. But a formal union vote signals something more structural: workers want a permanent seat at the table when these decisions get made, not just the ability to send angry emails.

For the broader AI regulation debate, this matters. If internal pressure from within labs grows into organized labor, it creates a new accountability mechanism that policymakers haven't fully accounted for.

Read the full story →

Ashley MacIsaac, a three-time Juno award-winning Canadian fiddle player, has filed a $1.5 million civil lawsuit against Google, alleging that Google's AI Overview feature falsely and publicly identified him as a sex offender.

The inaccurate AI-generated summary of his life and career reportedly led to a concert cancellation. MacIsaac says the defamatory information appeared prominently in search results, giving it the kind of credibility that a random blog post wouldn't have. When Google's own interface presents false information as fact, the reputational damage is severe and fast.

This case is one of the clearest examples yet of AI hallucinations causing direct, quantifiable harm to a real person. It's no longer a theoretical risk. The $1.5M figure will almost certainly get attention from Google's legal team - and from other AI developers watching to see how liability shakes out in court.

If you're building anything that surfaces AI-generated summaries about real people, this case is required reading. The legal precedent being set here could reshape how search and AI Overview products handle biographical content entirely.

Read the full story →

AI chip maker Cerebras is heading toward what could be one of the biggest tech IPOs of 2026, with a target valuation of $26.6 billion or more. And its secret weapon is a deep, ongoing relationship with OpenAI.

Cerebras makes the Wafer Scale Engine - a chip the size of an entire silicon wafer that's purpose-built for AI inference at speed. Its close partnership with OpenAI gives it a level of revenue predictability that most chip startups can only dream of, and it's exactly the kind of story public market investors are hungry for right now.

The AI infrastructure investment wave is far from over. With Nvidia still dominating training workloads, there's real appetite for alternatives in inference - which is where Cerebras is positioning itself. A successful IPO here would be a strong signal that the AI chip market has room for more than one winner.

On the enterprise side, both Anthropic and OpenAI are also separately launching joint ventures with asset managers to push harder into enterprise AI services - a sign that the race to own corporate AI spend is intensifying across the board.

Read the full story →

Colin Angle, the engineer behind the Roomba and the man who put 50 million household robots into people's homes, has launched a new company called Familiar Machines & Magic - and its first product is nothing like a vacuum cleaner.

The "Familiar" is a furry companion robot designed to interact autonomously with family members. Unlike the Roomba's single-minded goal of clean floors, this device is built around presence and personality. It's intended to be something you relate to, not just something that runs in the background.

The companion robotics space has been heating up for a couple of years, but it's mostly been startups without proven consumer hardware track records. Angle brings something different: he actually knows how to get robots into homes at scale. The Roomba succeeded because it was useful and unobtrusive. Whether a social companion robot can thread that same needle is the big open question.

If you're building AI-powered apps or companion experiences and thinking about your landing page, 60sec.site is worth checking out - it's an AI website builder that can get you from idea to live site in about a minute.

Read the full story →

Quietly but meaningfully, Google has added event-driven webhooks to the Gemini API - specifically for Batch API jobs, Deep Research tasks, and video generation.

If you've built anything on top of long-running AI jobs, you know the pain of polling: your app repeatedly asks "are you done yet?" until the task completes. Webhooks flip this around - the API calls you when it's done. The new system includes built-in security, retry guarantees, and two configuration modes.

For developer tools that rely on batch processing or async video generation, this is a real quality-of-life upgrade. Less wasted API calls, cleaner code, lower costs. It's the kind of infrastructure improvement that doesn't make headlines but makes daily development noticeably less annoying.

Track all your Gemini API usage and token costs with our Token Calculator at Daily Inference.

Read the full story →

The answer is B - the Cerebras WSE (Wafer Scale Engine) uses a single chip the size of an entire silicon wafer. While a standard GPU might be a few centimeters across, the WSE covers the entire wafer surface - roughly the size of an iPad. This gives it an enormous number of on-chip cores and a massive amount of on-chip memory bandwidth, making it exceptionally fast for AI inference workloads without the bottleneck of moving data between multiple chips. It's a genuinely unusual engineering approach, and it's a big part of why OpenAI has been happy to use it.

The MacIsaac lawsuit raises a real question: have you ever personally seen an AI-generated summary get something badly wrong about a real person or event? Hit reply and tell me what you saw - I read every response and the answers are often more surprising than the stories we cover.

That's all for today. For more AI coverage, visit dailyinference.com - see you tomorrow!