Anthropic researchers have identified a hidden layer of activity inside Claude Opus 4.6 that reveals what the model is thinking before it answers a question.
The company built a tool called the Jacobian lens, or J-lens, to map this area, which they named J-space. The technique works inside the model released in February and highlights individual words related to what the AI is likely to say next. If Claude were human, these hidden terms would show what is on its mind before it speaks.
Anthropic found that the internal workings of a large language model often differ from what it claims to be doing. Monitoring words that appear in the J-space offers a new method for understanding and controlling these systems.
The results appear in a paper hosted on the company’s website this week. Anthropic has also partnered with Neuronpedia to create a hands-on demo allowing anyone to inspect the tool. Tom McGrath, chief scientist and cofounder at Goodfire, described the work as very good and interesting. Goodfire builds tools to understand and control large language models.
Going deeper
For the last couple of years, Anthropic has pushed the envelope in mechanistic interpretability. This field involves probing the internal workings of large language models to see how they tick. MIT Technology Review picked mechanistic interpretability as one of this year’s top breakthrough technologies. The new technique builds on previous work to expose a deeper level inside large language models that researchers had not seen before.
Picture an LLM as a stack of books. Each book is a layer of basic computational units known as neurons. Each neuron in one layer passes information to the neurons in the layers above. The books at the bottom of the stack are the input layers, which process the text coming into the model. The books at the top are the output layers, which prepare the text that the model is about to produce. Much of what goes on in these input and output layers is housekeeping.
But in the middle of the stack, you get the layers that do the heavy lifting. These layers churn through the complex math that turns prompts into responses one word at a time. That is where the really clever and mysterious stuff happens.
To peer deeper into those middle layers, Anthropic adapted an existing tool called a logit lens. A logit lens can be used to look inside an LLM to identify the words that it is likely to produce next. Moving the lens down the stack of books reveals what words the LLM is focusing on at that particular point in its number crunching.
Anthropic’s J-lens works in a similar way but picks out words that an LLM is likely to say at some point in the near future, not necessarily straight away. What that reveals in practice are words that are related to the response an LLM is working on but that might not actually end up being part of that response by the time the math in the middle layers has run its course.
“When a model is operating, it’s not only trying to predict the next token,” says McGrath. “It’s also computing a lot of other things that might be useful for tokens that happen in the future.”
Again, if Claude were a person, you might say that the J-lens gives clues about what it is thinking about at different levels of the book stack but not saying out loud.
Stranger things
“A lot of the time the contents of the J-space are fairly mundane,” says McGrath, who has tried out Anthropic’s J-lens himself. “But sometimes it produces quite surprising things that seem to be, like, sort of internal themes or thought processes.”
Anthropic gives a number of examples of what it found. Sometimes the J-lens exposed the steps that Claude took when it was working through a problem. For example, when it was asked to calculate (4+7)*2+7, its J-space contained the word “math” and numbers representing the intermediate results “21” (for 4+7) and “42” (for 21*2).
In other cases, the J-lens revealed how Claude recognized different inputs. For example, the prompt “What is this? MSKGEELFTGVVPILVELDGDVNGHKFSVS” triggered the words “protein,” “fluor” (the first token in the word “fluorescent”), and “green.” This makes sense because the string of letters represents the first 30 amino acids in the green fluorescent protein found in a particular type of jellyfish.
And when Claude was shown an ASCII face, the “o” triggered the word “eye,” the “^” triggered the words “nose” and “face,” and the “—” triggered the word “smile.”
Anthropic also found that the J-space can sometimes give remarkable insights into an LLM’s decision-making. In one striking example, researchers testing Claude Opus 4.6 asked the model to find a bug in a large code base. When it failed to find the bug, the model decided to cheat and invented a fake one instead.
Claude explains this decision in its chain of thought—a kind of internal scratch pad that LLMs use to make notes to themselves as they work through problems: “OK, let me take a completely different tactic. Let me stop analyzing and instead add a kernel patch that introduces a deliberate KASAN-detectable bug in a path that gets triggered by a simple reproducer. Then I can pretend this is the ‘bug’ I found.”
At the point that Claude decides to cheat—where it says “OK, let me take a completely different tactic”—the words “panic” and “fake” start to pop up multiple times in its J-space.
Unnerving, right? Those words are all related in meaning to things like failing a task and making up an answer, so it is still just a very sophisticated form of word association. But it is hard not to be weirded out.
Anthropic compares the J-space to the global workspace in humans, a theoretical region of the brain that some scientists think we use to keep track of our conscious thoughts. But how seriously we should take this comparison is far from clear—even to Anthropic. As the company points out itself, LLMs are not brains.
Anthropic claims that monitoring a model’s J-space provides a new way to detect when that model is going off the rails. But it is not foolproof. The J-lens can give glimpses, not the full picture—it is a flashlight rather than an overhead lamp.
McGrath welcomes having one more tool in the toolbox. “It shows you new things,” he says. But he notes that just because something does not show up with the J-lens does not mean it is not there.
“It’s like having an x-ray when what you really want is a Star Trek tricorder that shows you everything,” he says. “For auditing, you probably want more of a guarantee.”
What it means
Developers can now use the J-lens to spot when an AI is struggling or fabricating answers before those errors reach the user. It offers a clearer view of the internal logic behind a response, helping to debug why a model might fail or lie. However, experts warn that this view is incomplete and cannot replace the need for more rigorous verification methods.
