Behind every exciting discovery, some real people worked long and hard at it. Anthropic J-Space women in AI Most people who read about the discovery will never even learn those names of the creators. In July, a company called Anthropic did discover something like this. They call it “J-Space.” Anthropic J-space women in AI helped make it possible.
Some women have worked on the Anthropic project doing hard, important work for a long time. But almost nobody notices them because their work is too technical to be trending news or viral posts.
Before introducing those women, let’s talk about the actual discovery they made.
The J-Space Breakthrough
Right now, your brain is doing an almost absurd amount of machinery you’ll never notice.
Most of the time, you have zero awareness of balancing your body, air in your lungs, or light hitting your eyes into recognizable shapes and figures.
Scientists call that sliver conscious access. It’s the tiny repeatable part of everything your brain is up to.
What Anthropic’s researchers found is that something surprisingly similar seems to happen inside their AI models. Buried in the layers of computation that make up a system like Claude, ChatGPT, or Gemini there’s a small, specific area they’re calling “J-Space” that holds the concepts the model is actively thinking through.
It’s a tiny fraction of everything happening inside the model at any given moment, but it’s the part that matters most for actual reasoning. It’s the difference between an AI casually finishing a sentence and one working through a hard problem step by step.
With this, we are not saying AI is conscious: the researchers themselves are careful about that distinction and it’s an important one. What they are claiming is something almost as interesting: that if you build a system smart enough to reason, something resembling a mental workspace might show up whether you designed it to or not.
The Women Shaping AI Safety
Amanda Askell, Anthropic
Amanda Askell trained as a philosopher, not an engineer. She has a PhD from NYU with a thesis on infinite ethics and previously worked on AI policy at OpenAI before joining Anthropic in 2021. Today she leads Anthropic’s work on model character and alignment, effectively the team responsible for how Claude behaves, reasons about tricky situations, and holds (or doesn’t hold) to its stated values. The Wall Street Journal put it simply: her job is teaching an AI how to be good.
Victoria Krakovna, Google DeepMind
Krakovna’s path ran through a PhD in statistics at Harvard, focused on building interpretable models, long before interpretability was a term outside academic circles. She’s spent years on some of AI safety’s thorniest open problems: deceptive alignment, specification gaming, and the ways a system can technically do what you asked while missing what you meant. She also co-founded the Future of Life Institute, one of the organizations that’s spent over a decade pushing the industry to take these risks seriously before they were fashionable to worry about.
Stella Biderman, EleutherAI
Biderman runs EleutherAI, a research institute built on a genuinely unusual premise: that understanding how large language models work shouldn’t be locked inside the handful of companies that build them. Her own research sits squarely in interpretability and part of her role has been fighting to keep that kind of research open to academics and independent researchers who’ll never have a corporate lab’s budget.
Why This Matters Beyond the Headlines
It’s easy to file all this under tech news and move on. That would be a mistake, and here’s why.
Interpretability and alignment research isn’t really about satisfying curiosity over how a model works. It’s about who gets to decide what these systems value, what they’re allowed to do, and what happens when they get something wrong at a scale that affects millions of people at once.
Askell’s work shapes what good behavior even means for a system that talks to more people in a day than most humans do in a lifetime. Krakovna’s work is, in plain terms, about making sure powerful systems don’t quietly pursue the wrong goal while appearing to do exactly what was asked. Biderman’s work decides whether that understanding stays locked inside a few companies or stays available to the researchers, regulators, and academics who’ll need it to hold those companies accountable.
This is the interesting part. Askell didn’t start as an engineer. Krakovna’s early work was in statistics, not machine learning as it’s taught now. Biderman built her career arguing that powerful research shouldn’t stay locked behind a handful of corporate doors. None of them fit the image most people still picture when they hear AI researchers.
Today, various tech institutions such as Alt School Africa are helping to bridge the gender gap by providing women with opportunities to delve into tech, learn and build a career in by providing dedicated scholarships to teach and launch African women into frontier engineering fields.
There’s a lesson here for any woman building a career in tech or otherwise. The most consequential work rarely announces itself as consequential while you’re doing it. Nobody was writing headlines about interpretability research five years ago. The women did it anyway. The theory was that it mattered to understand how something worked before it became powerful enough to no longer be optional. That’s a bet worth making in any field not just this one.
So next time you see an AI breakthrough headline or any kind of breakthrough headline, it’s worth asking the question this piece started with: who actually built this thing, and how long were they doing it before anyone noticed? Their names were always worth knowing. Now you know a few.
Thank you for reading! Want a natural, radiant look even without makeup? Read our article on how to Achieve a Natural, Radiant Look without makeup.
Staff writer at Women Digest covering beauty, fashion, wellness and life.