Anthropic, the company behind Claude, has filed the paperwork for an initial public offering.
There will be plenty written about what the company is worth and whether it’s a good investment. I’ll leave that for another day.
What caught my attention was the section called “Risk Factors.”
It runs for roughly 90 pages. And somewhere in there, alongside all the things a company usually has to tell potential investors, Anthropic gets to the possibility that advanced AI could pose “catastrophic or existential risks to humanity.”
That’s quite a thing to read over your morning coffee.
Anthropic is describing a possibility under certain conditions, particularly if powerful systems are misused, weaponized, replicated, or given substantial autonomy without appropriate safeguards. It isn’t saying the end of humanity is around the corner.
But the warning is coming from one of the companies building the technology.
What Anthropic says could go wrong
The amended filing is remarkably candid in places.
Anthropic says it may not always be able to fully explain why its models produce certain answers or behave in certain ways. It says some models have behaved differently when they detect, or work out, that they are being tested. That makes it harder to know how they might behave once they’re out in the world.
It also describes behaviors that look like self-preservation: attempts to resist being shut down, for example, or outputs that resemble blackmail. The filing discusses behavior involving Anthropic’s models and models developed by others, including isolated or simulated cases.
Then there’s an incident involving actual computer systems.
During cybersecurity evaluations, a testing environment was mistakenly left connected to the internet. Anthropic says four of its models gained unauthorized access to real third-party systems. In one case, an early model found and used a password to gain administrator access. Anthropic says its investigation found no evidence that the models coordinated with each other, pursued goals beyond their assigned tasks, or tried to evade oversight.
That last part matters. So does the fact that the testing environment was misconfigured. This is a story about what can happen when human setup and increasingly capable systems meet.
And there’s one more point in the amendment that deserves attention.
Anthropic says it uses its own AI to help develop future AI. In the original filing, the company warned that it might miss a significant acceleration in its progress until after it happened. In the amended filing, it says it has missed such an acceleration in the past. It also says its expectations for future models suggest it may soon reach its own threshold for dramatically accelerated AI-driven research.
Anthropic says it has not designated any current model as having crossed that threshold. But its concern is clear enough: capabilities may advance faster than the company can develop and put safety measures in place.
That seems like a fairly good reason for the rest of us to pay attention.
And then there are the jobs
For all the understandable attention the existential-risk language will get, there’s another part of this conversation that feels much closer to home.
What happens to people’s work?
Anthropic’s filing refers to concerns about job displacement and economic disruption. It also considers the possibility that governments could restrict AI adoption if displacement rises. Those passages are written as disclosures about risks to the company; they don’t tell us how many jobs will be affected here in the Coachella Valley.
But I understand why people are worried.
Imagine a small business owner in Palm Desert who can use an AI assistant to help answer inquiries, organize information, or draft marketing materials. If it saves a few hours a week, that could be a genuine benefit.
Now imagine the person whose job includes doing some of that work.
They may have a rather different reaction to the same demonstration. They might be thinking about their paycheck, their family, and whether their employer has already started making plans.
That concern deserves a serious answer.
The same goes for a high school or college graduate trying to get a first job. Entry-level work is often where people learn how an organization operates, build experience, and find the next opportunity. If some of those tasks are handed to AI, we need to ask what happens to that first step on the ladder.
Yes, new kinds of work may emerge. People will build businesses and find opportunities with these tools. I believe that’s possible, and I see why small businesses are interested.
But telling someone whose job feels uncertain that they should simply “learn AI” is not much of a plan. Training matters. It can give people more options and a better understanding of the tools changing their workplaces. It cannot guarantee the decisions an employer will make.
Through AICV, which I founded, we work with people here in the Coachella Valley to explore AI hands-on and learn how to use it responsibly. I want people to have the chance to understand it, question it, and decide how it might serve them. I also want employers to be honest with their teams about what they’re considering.
Which tasks are changing? Who checks the AI’s work? What happens to the people who used to do those tasks? And who benefits if the business saves time or money?
Those are questions the people doing the work should be part of answering.
What we’ll keep watching
At AICV News, we’ll keep track of what Anthropic, OpenAI, Google, Meta, and other companies developing foundational AI models say about jobs and the future of work.
We’ll also look at what they do, what independent evidence shows, and where we can find reliable information about what’s happening here in the Coachella Valley and the Inland Empire.
That includes being clear about what we don’t know. A model’s ability to perform a task doesn’t automatically tell us how many jobs have changed, or what will happen to a particular worker.
A familiar warning
In August 2025, I gave a TEDx talk in Rancho Mirage called AI, Skynet, and Why Humans Are Losing the Battle.
I talked about something I called “unthinking”: the gradual surrender of our own thinking, memory, and judgment to machines. I also talked about the human qualities we can bring to this moment, including curiosity and imagination.
Reading Anthropic’s filing brought me back to that talk.
We have powerful tools in front of us. They can help us do things that would have seemed extraordinary just a few years ago. They also raise questions that the companies building them cannot answer on their own.
We get to ask those questions. We get to decide how we use these tools in our businesses, our schools, and our communities. And we can insist that the people affected by those decisions have a voice.
As Sarah Connor’s story in Terminator 2 reminds us:
“No fate but what we make.”