By Chuck Gallagher — Business Ethics Keynote Speaker and Trainer
TL;DR: Chuck Gallagher, AI ethics speaker and author, examines why the people building the world’s most advanced AI systems are now organizing to protect themselves from the very tools they helped create.
When AI Builders Ask for Protection From Their Own Work
At noon on Thursday, roughly a hundred people stood on a strip of grass in Mountain View, the Googleplex on one side and the visitor center on the other. Matching black shirts. White cardboard signs lettered in black marker. “Googlers for Job Security.” Someone unrolled a banner covered with names — 4,500 of them, coworkers who signed a petition asking the company for some assurance they would still be employed next quarter.
These were not outsiders protesting the AI boom from the sidewalk. These were the people building it.
Sit with that a minute.
What Were Google’s Workers Actually Asking For?
Not much, as demands go. Voluntary exit offers before layoffs. Guaranteed severance standards. An end to performance rating quotas. The option to take severance as extended paid leave. Parul Koul, a Google software engineer and president of the Alphabet Workers Union, told the crowd that workers want the security to do their best work instead of spending their days in an environment driven by fear, pitted against colleagues, never sure how much longer the job lasts. The union has about 1,400 members. The petition carried 4,500 names.
About twenty workers carried the petition to the offices of Google Cloud CEO Thomas Kurian and senior vice presidents Rick Osterloh and Nick Fox. Nobody was in. They left copies outside the doors. At Sundar Pichai’s office, a member of his team accepted it. Google did not comment. As an AI ethics speaker and author, I have watched organizations arrive at this exact spot. The people closest to the work name the consequence before the people signing the checks will. That is not a Google problem. It is a human problem, older than software.
Is This Only Happening at Google?
No. And the pattern outside Google is sharper.
In April, Reuters reported that Meta was installing software on U.S. employees’ work computers to capture mouse movements, keystrokes, and screenshots across work applications, all of it feeding AI training. The program is called the Model Capability Initiative. The internal memo told staff this was where every Meta employee could help the company’s models improve simply by doing their daily work. The goal is agents that complete computer tasks on their own. More than 1,600 workers signed a petition against it. In June, Meta said employees could pause the collection for thirty minutes at a time.
Now read the sequence. Capture the keystrokes. Train the agent. Then, starting May 20, lay off roughly 8,000 people — about 10% of the workforce — while moving some 7,000 others onto AI-focused teams. Meta framed the cuts as the way to fund an AI buildout guided at $115 billion to $135 billion in capital spending this year. Fourth quarter revenue was a record $59.89 billion.
Then it moved to a courtroom. On July 13, twenty-six Meta employees sued in federal court in Oakland. They allege Meta used internal AI systems, keystroke and activity-monitoring data, AI token-usage dashboards, and algorithmically assisted performance rankings to decide who got cut — and that those scores, by design, cannot be accumulated by someone on protected medical or parental leave. Separations were set to begin July 22. Meta says the claims lack merit and that workforce decisions were made by people, not AI. Allegations, not findings. A court will sort it out. But look at what is alleged. The measuring stick was the product.
Oracle tells a plainer version. In April, more than 600 laid-off Oracle workers signed a letter asking for better severance and longer healthcare. Their claim was that they had been used to train AI systems and then let go. Oracle replied that it would accept only individual letters. They sent those. Nothing happened. At Google DeepMind, about 300 UK workers joined the Communication Workers Union, and in April, 98% voted to seek recognition for roughly 1,000 staff, largely over military AI contracts.
The executives are not hiding the math. Salesforce CEO Marc Benioff said his support organization went from 9,000 heads to about 5,000 because, in his words, he needs less heads. Support costs fell 17%. Amazon cut 14,000 corporate roles in October and 16,000 more in January, after CEO Andy Jassy told employees that generative AI and agents would mean fewer people doing some of the jobs done today. Stanford’s 2026 AI Index found employment for software developers ages 22 to 25 down nearly 20% since 2024.
Why Does This Keep Taking the Same Shape?
Here is the pattern, stripped down. A worker is asked to contribute to a system. The contribution is framed as routine — just do your daily work. The system improves. The worker is then measured by the system. The measurement decides who stays. Nobody in that chain has to be a villain. That is what makes it worth studying.
Trust me. The worst ethical failures inside American companies never announce themselves. They arrive as reasonable next steps, each one defensible alone, and by the time you see the whole staircase you are standing at the bottom of it. I know that ground personally. Every choice has a consequence. It does not care whether the choice felt small when you made it.
Notice how modest that Mountain View petition is. Every demand on it asks for the same thing: make the consequence visible before it lands. Not stop it. Say it out loud.
What Does Honest AI Leadership Look Like Right Now?
It looks like disclosure that costs you something. As an AI ethics speaker and author, I tell boardrooms the same thing every time. If an AI-assisted score helps decide who eats next year, tell your people it exists. Tell them what it measures. Tell them what it cannot measure — that it will not see the engineer on medical leave, the parent on twelve-week absence, the quiet one who fixes everybody else’s mistakes and never logs a token. Consent that requires ignorance is not consent. It is just quiet.
The Googlers on the grass were not asking anyone to stop the machine. They built the machine. They were asking to be told the truth about what it is pointed at. That is a low bar. Somebody ought to clear it.
Frequently Asked Questions
Why are Google employees protesting AI if they build it?
The rally outside Google’s Mountain View headquarters was about job security, not opposition to AI itself. Workers delivered a petition signed by 4,500 employees asking for voluntary exits before layoffs, guaranteed severance standards, and an end to performance rating quotas. Their concern is that the efficiency gains they are producing are being converted into headcount reductions without any corresponding protections for the people producing them.
Can a company legally use AI to decide who gets laid off?
That question is currently in front of a federal court. Twenty-six Meta employees sued in Oakland on July 13, alleging the company used internal AI systems, activity monitoring, AI token-usage dashboards, and algorithmically assisted rankings to select workers for layoffs, disproportionately hitting people on medical and parental leave. Meta says the claims lack merit and that its decisions were made by people, not AI. Nothing has been proven, but the case is a preview of where this fight is headed.
Is it ethical to have employees train the AI that may replace them?
It depends almost entirely on disclosure. As an AI ethics speaker and author, I draw the line at whether a person can make an informed choice about their own participation. Capturing keystrokes and screen activity to train agents is a defensible business decision. Doing it while framing it as routine work, without telling employees the agent is aimed at tasks they currently perform, is not.
How many tech jobs have actually been cut because of AI?
Attribution is genuinely murky, because companies rarely label a cut as AI-driven. What is documented is the scale: close to 400,000 tech workers have been laid off since the start of 2025, Amazon has eliminated more than 30,000 corporate roles across two rounds, and Meta cut about 8,000 in May. Stanford’s 2026 AI Index found employment for software developers ages 22 to 25 down nearly 20% since 2024.
What should a company do before deploying AI in workforce decisions?
Disclose the system before you use it, not after someone sues. Tell employees what is being measured, publish what the metric cannot see, and build a human review that can override the score for people on protected leave or accommodation. If you would not be comfortable explaining the method to the workforce in a room, you should not be comfortable using it in a spreadsheet.
Take the Next Step
Every organization deploying AI in workforce decisions is making a choice right now, whether it is naming that choice or not. Chuck Gallagher works with boards, executive teams, and conferences on exactly this problem — how to adopt powerful technology without quietly outsourcing your ethics to it, and how to tell your people the truth before the consequence arrives instead of after. If your leadership team is having this conversation, or avoiding it, bring in someone who has lived the cost of a small, reasonable-seeming decision. Learn more or book a conversation at ChuckGallagher.com.
Five Questions for Reflection
1. If your company deployed a system that measured your daily work, would you want to know what it could not see about you? Would you ask?
2. Where is the line between asking employees to use a tool and asking them to build their own replacement — and who in your organization is authorized to draw it?
3. The Google petition asked for disclosure, not protection from change. Why is disclosure so often the harder ask?
4. Think of a decision your organization made that seemed reasonable at each step. Looking back, where was the moment someone could have named the consequence out loud?
5. If an AI-assisted score helped decide who stays at your company next year, could you explain the method to the people it ranked? If not, what does that tell you?
