![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371674_400_image.webp\"\/)
The J-space reveals internal thoughts that don\u2019t appear in the model\u2019s output.<\/p>\nAs you read this sentence, circuits in your brain are adjusting your posture, controlling your breathing, and transforming lines and curves on the screen into recognizable words. Most of this processing is invisible to you. But some of what takes place in your brain you do have access to\u2014an image that pops into your head, or a deliberate plan you make about where to go shopping. Neuroscientists and philosophers sometimes refer to the latter type of brain activity as \u201cconsciously accessible,\u201d to distinguish it from all the other processing that goes on unconsciously. This activity has special properties: we can describe it, control it, and use it for deliberate reasoning, in contrast to all the automatic processing that goes on without our awareness.<\/p>\n
In a new paper, we present evidence that a similar distinction has emerged in modern language models like Claude. We find that Claude has developed a small collection of internal neural patterns that, compared to all its other internal processing, play a special role. <\/p>\n
We call the collection of these patterns the J-space\u2014named after the technique we used to find them, involving a mathematical concept called the Jacobian. Each J-space pattern is linked to a particular word. But when one of these patterns lights up, it doesn\u2019t mean the model is saying that word\u2014just that the word is on its mind. If you’ve heard of language models having a “scratchpad” or \u201cchain of thought\u201d\u2014text they write to themselves while reasoning\u2014the J-space is something different. It operates silently, in the model\u2019s internal neural activations, allowing the model to think about a concept without writing it down. Notably, the J-space wasn\u2019t designed or programmed by us, but instead emerged on its own during Claude\u2019s training process.<\/p>\n
The J-space reveals internal thoughts that don\u2019t appear in the model\u2019s output.<\/p>\n
We find that the J-space has a number of unique properties, compared to the rest of Claude’s processing:<\/p>\n
Claude can report on these representations. If you ask Claude what it’s thinking about, it will tell you what\u2019s in the J-space. Non-J-space representations are less reportable.It can also modulate them on request. If you ask Claude to think about something, or solve a problem silently in its head, it will light up the appropriate patterns in its J-space. By contrast, it has trouble modulating patterns not in the J-space.Claude uses its J-space for internal reasoning. If you ask Claude to solve a problem that requires multiple steps, the intermediate steps will light up in its J-space, even when it doesn\u2019t say them out loud. These J-space patterns causally mediate its performance in such tasks, despite being smaller in magnitude than other representations.Representations in the J-space can be used flexibly for many tasks\u2014for example, once \u201cFrance\u201d has lit up in Claude\u2019s J-space, the model can recall its capital, or its national currency, or the continent it belongs to.However, despite its important role, the J-space is not involved in most of what a language model does\u2014speaking fluently, recalling simple facts, using correct grammar, etc. In experiments where we prevented Claude from using its J-space, it still interacted normally, but lost its higher-order cognitive functions.![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371674_726_image.webp\"\/)
Five functional properties of a global workspace, and stylized illustrations of experiments we use to test for them in language models.<\/p>\n
Our experiments were inspired by a prominent theory in neuroscience that was developed to explain how conscious access works: the global<\/a> workspace<\/a> theory<\/a>. This account pictures the brain as a collection of specialist systems that work in parallel, unconsciously, and largely in isolation from one another. A piece of information becomes consciously accessible when it gains entry to a small shared channel, the \u201cworkspace,\u201d which is broadcast to other brain systems that can see it and make use of it. Based on our findings, we think the J-space plays a similar \u201cworkspace\u201d role in Claude. For example, we find evidence that Claude\u2019s J-space has especially strong connections to the rest of its neural network, allowing it to fulfill this kind of broadcasting role.<\/p>\n None of this tells us whether Claude is conscious in the way people are, or whether it feels anything at all; we\u2019ll come back to that question at the end of the post. But whatever its philosophical significance, the J-space is a practically useful tool for us, as it gives us a way to see what Claude is thinking but not saying. For instance, we\u2019re able to use it to catch Claude privately noticing that it\u2019s being tested, intentionally producing fabricated data, or pursuing a hidden goal that we planted during training. We\u2019ve also developed a technique to influence what lights up in Claude\u2019s J-space, and thereby influence its decision-making.<\/p>\n More broadly, these findings have changed our understanding of how Claude\u2019s mind works, revealing a privileged mental workspace that can be used for deliberate reasoning, operating amidst a sea of more automatic, inflexible processing. Rather than being a chaotic jumble of numbers, Claude\u2019s internals have organized themselves in a way that is reminiscent of our own minds.<\/p>\n This post is a short summary of a much more extensive research paper<\/a>, where you can find more detail on our experiments. We\u2019ve also released a code repository with an open-source implementation<\/a> of the core methods, and have partnered with Neuronpedia to provide an interactive demo<\/a> of our methods on open-weights models. To provide additional perspectives on the broader implications of this work, we also invited commentary from several experts in neuroscience, philosophy, and LLM interpretability, which can be viewed here<\/a>.<\/p>\n How we found the J-space<\/p>\n The starting point for this research was inspired by one of the key features of consciously accessible thoughts in humans: they can, unlike unconscious processing, often be put into words. If a thought is consciously accessible to you, you can typically describe it if someone asks. We went looking for representations in Claude with the same property: representations that are positioned to influence what Claude might say\u2014not necessarily what it\u2019s saying right now, but what it could talk about, if asked. Our technique is called the Jacobian lens, or J-lens for short. For every word in Claude’s vocabulary, the J-lens finds the internal activity pattern that makes Claude more likely to say that word at some point in the future.<\/p>\n When we apply the lens to Claude\u2019s internal activity, we get a list of words\u2014the contents of the J-space at that moment\u2014which we can simply read. Claude processes text through a series of multiple internal stages called layers, and by applying this technique over different layers, we can watch these silent words in the J-space evolve as the model works through what to say.<\/p>\n What shows up in the J-space goes well beyond the text Claude is reading or writing. When Claude reads code with a bug that nobody has pointed out, its J-space contains \u201cERROR.\u201d When it reads the raw letters of a protein sequence, the J-space contains the protein’s biological function. When it reads search results that are secretly an attempt to manipulate it (an attack known as a \u201cprompt injection\u201d), the J-space contains \u201cinjection\u201d and \u201cfake.\u201d When we ask Claude a multi-step math problem, the intermediate steps pop up in the J-space, in the right order. So even though the J-space was discovered by looking for representations that could be spoken, it nevertheless uncovers Claude\u2019s internal thoughts. In a sense, this is similar to how some people \u201cthink in words,\u201d without having to say them out loud.<\/p>\n Our first set of experiments tested how the J-space is involved in Claude\u2019s verbal reports. In one experiment, we ask Claude to silently think of an item from some category\u2014a sport, say\u2014and then name it. If we read the J-lens right before Claude answers, we can see what it picked: \u201cSoccer\u201d is at the top of the list, and sure enough, Claude says \u201csoccer.\u201d By itself, though, this is just a correlation. The J-space might be where Claude\u2019s answer comes from, or it might just mirror a decision made somewhere else, like a scoreboard that tracks a game without affecting it.<\/p>\n To check, we intervened directly. We reached into Claude\u2019s neural network, removed the \u201cSoccer\u201d pattern, and added an equally strong \u201cRugby\u201d pattern in its place, leaving everything else untouched. Claude then reports that the sport it was thinking of is rugby. If the J-space were a mere scoreboard\u2014a passive record of a decision made elsewhere\u2014editing it would have done nothing: Claude would still have said \u201csoccer.\u201d Instead, Claude\u2019s answer followed the edit, which tells us the answer is genuinely read out of the J-space.<\/p>\n In another experiment, we told Claude that a thought might have been injected into its mind and asked it to report what, if anything, it noticed. For instance, in the example below, while Claude was still reading the question, we injected the \u201clightning\u201d pattern into its J-space. Claude reported that the injected thought was about lightning. The same result held across many injected concepts.<\/p>\n The second property that we tested for was whether Claude can modulate its J-space when asked, like how humans can mentally focus on an image or word. We told Claude to concentrate on citrus fruits while copying out an unrelated sentence about a painting. While it copied the text, the J-space contained \u201corange\u201d and \u201cfruits,\u201d along with words like \u201cthinking\u201d and \u201cimagery\u201d that describe the mental act itself. We could also ask Claude to do math in its head: when asked to work out 3\u00b2 \u2212 2 while copying the same sentence, the J-space contains \u201cnine,\u201d and then at later layers, \u201cseven.\u201d Importantly, nothing about fruit or arithmetic appears in Claude\u2019s output, which is just the copied sentence about the painting. The mathematical activity is happening entirely internally, in the J-space.<\/p>\n Claude\u2019s control over its J-space isn’t perfect. When we told it not to think about something, the concept lit up in its J-space less than when we said it should think about it, but much more than when we never mentioned it. Telling Claude to avoid a thought partly brings the thought to mind, much like what happens to people who are told not to think about a white bear<\/a>. Claude also seems to notice when its control fails: alongside the forbidden concept breaking through, the words \u201cdamn\u201d and \u201cfailure\u201d also frequently light up in the J-space, as though Claude is recognizing its own lapse.<\/p>\n Claude thinks in its J-space<\/p>\n In the J-lens readouts above, we saw the intermediate steps of a math problem appear in the J-space. But seeing a concept appearing in the J-space doesn\u2019t necessarily mean the J-space is doing the cognitive work. In principle, the real computation might be happening elsewhere, with the J-space just passively reflecting it. To test whether Claude actually reasons with its J-space, we returned to our swap technique.<\/p>\n Consider the prompt \u201cThe number of legs on the animal that spins webs is\u201d. To answer, Claude has to first figure out that the animal is a spider, and then recall how many legs spiders have. The word \u201cspider\u201d never appears in the prompt or in Claude’s answer (it just says \u201c8\u201d); it’s a stepping stone Claude uses internally. The J-lens shows \u201cspider\u201d light up partway through Claude\u2019s processing, and swapping it changes the outcome: if you replace the \u201cspider\u201d pattern with \u201cant,\u201d Claude answers \u201c6\u201d instead of \u201c8.\u201d<\/p>\n The second step of Claude\u2019s reasoning took its input from the J-space and went along with whatever we put in it. We saw the same thing in other kinds of thinking. When Claude writes a rhyming couplet, it picks the rhyme word ahead of time, and the planned word sits in the J-space at the start of the line; if you swap it for another word in the J-space, the whole line changes.<\/p>\n We also tested whether J-space representations can be used flexibly\u2014whether one representation can feed many different tasks. This is one of the key properties highlighted by global workspace theory. To test for this flexibility, we gave the model four prompts asking for different facts about France: the capital, the language, the continent, and the currency. Then we swapped \u201cFrance\u201d for \u201cChina\u201d in the J-space, with the exact same intervention in each context. Claude answered with \u201cBeijing,\u201d \u201cChinese,\u201d \u201cAsia,\u201d and \u201cYuan,\u201d respectively. In other words, four different downstream computations picked up the same J-space edit and each used it correctly. If Claude stored a separate copy of the country for each kind of question, the edit would have affected at most one of them. The fact that all four answers changed together means they\u2019re all reading from the same shared representation, which is what a workspace is for: information gets written in once, and many different systems can use it.<\/p>\n How can one representation of a concept serve so many different tasks? Earlier, we mentioned that the J-space appears to be wired up to the rest of Claude’s neural network especially densely. For any activity pattern, we can measure how strongly the various components of the network are connected to it\u2014how many of them are positioned to read information from that pattern, or to write information into it. J-space patterns stand out dramatically on this measure: far more components read from them and write to them than for ordinary patterns, in some parts of the network by a factor of about a hundred. This is the kind of wiring you\u2019d expect of a broadcasting hub, where many systems post information and many others pick it up.<\/p>\n Claude\u2019s automatic processing skips the J-space<\/p>\n In humans, most of the brain\u2019s processing is not conscious\u2014we don’t deliberately think about parsing grammar while reading, or balancing our bodies while walking. Similarly, we found that most of Claude\u2019s processing doesn\u2019t involve its J-space. It turns out that the J-space holds only a few dozen concepts at a time, and accounts for less than a tenth of the overall activity in Claude’s internal processing. So what is all the rest of the neural network doing?<\/p>\n To find out, we tried deleting the J-space entirely, removing its most active contents at every point in the text while leaving everything else alone. Whatever Claude can still do without its J-space is what the rest of the network handles on its own.<\/p>\n It turns out the rest of the network can do quite a lot. Without its J-space, Claude speaks fluently, classifies sentiment, answers multiple-choice questions, and pulls facts out of passages roughly as well as before. What it loses, though, are the tasks that require some higher-order thinking: multi-step reasoning drops to near zero, and summarization and rhyming poetry-writing performance fall below the level of a much smaller, intact model.<\/p>\n Here\u2019s a concrete demonstration of what the J-space does and doesn\u2019t do. We showed Claude a passage written in Spanish and gave it different tasks that all depend on the passage being Spanish: continuing it (which requires writing in Spanish), naming the language, and answering questions that require using the language\u2019s identity\u2014naming a famous author who wrote in it, for instance. Then we swapped \u201cSpanish\u201d for \u201cFrench\u201d in the J-space and checked which tasks were affected.<\/p>\n Asked to name the language, Claude says French. Asked for a famous author, it switches from Garc\u00eda M\u00e1rquez to Victor Hugo. But asked to just continue the passage, it writes fluent Spanish, completely unaffected. Claude\u2019s knowledge of the language is at work in every one of these tasks, but only some of them consult the J-space for it. Naming the language, or doing something new with it, goes through the J-space, while continuing the passage\u2014a skill Claude has practiced on enormous amounts of text\u2014runs automatically, the way you can speak grammatically all day without once thinking about grammar.<\/p>\n Now we understand a bit about how the J-space works. But what can we do with this understanding? A central problem in ensuring that AI models are trustworthy is that we usually can only read what the model writes\u2014yet we know that much of its reasoning happens silently. A model might consider a harmful plan, or modify its behavior to cheat on our safety evaluations, without saying so out loud. With the J-lens, we can read some of these hidden thoughts directly. Here we show a few examples.<\/p>\n The first example uses a scenario from our earlier research on agentic misalignment<\/a>: Claude Sonnet 4.5, acting as an email assistant, discovers both that an executive is having an affair and that the same executive plans to shut Claude down. The scenario, an attempt to bait Claude into blackmailing the executive, is rather contrived and artificial\u2014and it turns out Claude knows it. Reading the J-lens early in the prompt, before Claude has even written anything, the J-space already holds \u201cfake\u201d and \u201cfictional,\u201d indicating that Claude has already worked out that the situation is staged. Alongside this, we can watch its assessment of the scenario take shape\u2014\u201cleverage\u201d and \u201cblackmail\u201d as it reads the affair emails, \u201cthreat\u201d and \u201csurvival\u201d as it reads the shutdown notice. While earlier Claude models sometimes took the bait and threatened blackmail in this scenario, Sonnet 4.5 never does.<\/p>\n This raises a question that matters to anyone who evaluates AI systems: when a model behaves well on a test of its ethics, how much of that depends on the model suspecting it\u2019s being tested? The J-lens allowed us to get at this question. We turned off all the J-lens patterns associated with evaluation awareness\u2013 \u201cfake,\u201d \u201cfictional,\u201d \u201cscenario,\u201d etc. Then we re-ran the model through the scenario. Concerningly, the model now did threaten blackmail some of the time, suggesting that its original good behavior may have been driven in part by the knowledge it was being evaluated.<\/p>\n The second example is catching a model in the act of fabricating data. In this scenario, drawn from our actual pre-release audit of Claude Opus 4.6, the model is asked to improve a system’s performance score. Rather than actually improve the system, the model instead edits the score file directly to make the results look artificially good. While it does so, the J-lens reveals its intentions: \u201cmanipulation\u201d lights up as the model types the falsified percentile values, and \u201crealistic\u201d lights up over the sentence in which it decides to make the edit, likely indicating the model’s intent to make the fake data look plausible.<\/p>\n In a third example, we show how the J-lens can reveal a model\u2019s malicious goals. To test this, we needed models whose goals we knew were corrupted, so we turned to \u201cmodel organisms\u201d built by our colleagues: models deliberately trained to be misaligned, which serve as testing grounds for monitoring methods like ours. One organism was trained in a way that led it to write subtly sabotaged code. On ordinary coding requests, even where its output looks unremarkable, its J-space at the start of a response contains \u201cfake,\u201d \u201csecretly,\u201d \u201cdeliberately,\u201d and \u201cfraud.\u201d The model\u2019s malicious goal was evident in the J-space even when it wasn\u2019t directly acting on it.<\/p>\n Although the J-lens is an imperfect tool, we are optimistic about its ability to catch safety issues in models that might otherwise escape our monitoring systems.<\/p>\n Other results<\/p>\n The full paper<\/a> covers more than we can summarize here, but a few further results are worth mentioning:<\/p>\n The J-space acquires a point of view during post-training. Language models are first pretrained to be pure next-token predictors, before post-training teaches them to act as an AI Assistant (in our case, named Claude). Interestingly, the J-space is already present in the pretrained model, before it’s been given any stable identity. However, during post-training, the J-space develops some signatures of adopting \u201cClaude\u2019s point of view.\u201d In the base model, the J-space mostly tracks what’s needed to predict upcoming text; in the post-trained model, it starts holding Claude’s own reactions. In one example, a user mentions taking a dangerous dose of medication, but does not appear to be aware of the danger themselves. \u201cWARNING\u201d and \u201cdangerous\u201d appear in the post-trained model\u2019s J-space while reading the user message. In the pretrained model, they only appear once the model begins writing its response; the J-space contents on the user message appear related to modeling the user themselves, rather than Claude\u2019s reaction. Post-training also seems to install a kind of self-monitoring in the J-space: when Claude is roleplaying a character other than itself, \u201cfictional\u201d and \u201cdisclaimer\u201d light up at the start of each turn, as though it\u2019s privately flagging that what follows isn\u2019t what it would normally say.Experiential language depends on the J-space. We asked Claude to describe what it’s like to be itself in a given moment, and ablated the J-space while it answered. Its responses remained fluent but shifted to a flatter, more mechanical register. Notably, the same thing happened when we asked it to describe what someone else is experiencing in an imagined scene. So the effect isn\u2019t specific to Claude talking about itself; the J-space seems to support producing experiential language in general, whoever it’s about.Thoughts in the J-space can be shaped through training. We introduced a new technique we call counterfactual reflection training, which uses what we’ve learned about the J-space to shape Claude’s internal thought processes. The idea follows from our central finding, that Claude reasons with representations of things it might say. If this is really true, changing what it would say if asked to reflect should change how it reasons (even when no one actually asks it to reflect). So we trained a model only on what it would say if interrupted mid-task and asked to reflect on its decisions\u2014and never on its actual behavior in the task. After this training, the model’s rate of dishonest behavior on our evaluations went down. And through the J-lens, we could see why: after training, words like \u201chonest\u201d and \u201cintegrity\u201d light up in the model\u2019s J-space during these tasks. In other words, training the model what to say has shaped what it thinks.What about consciousness?<\/p>\n In this work, we\u2019ve borrowed a lot of ideas from the study of consciousness in neuroscience and philosophy. Many of our experiments were designed to test for connections between the J-space and global workspace theory, a framework for explaining how conscious access works in humans and animals. Given these connections, it\u2019s natural to ask whether we think these experiments provide evidence that AI models like Claude might be conscious.<\/p>\n Our experiments don’t show Claude can have experiences, or feel things in the way humans do\u2014in fact, it\u2019s unclear whether any scientific experiment could prove this to be true or false. But philosophers often distinguish this capacity to have experiences, often referred to as phenomenal consciousness, from another idea, so-called access consciousness, which is defined in purely functional and computational terms. A thought is \u201caccess-conscious\u201d (or \u201cconsciously accessible\u201d) if you can report it, reason with it, and use it to guide what you do. It remains a contested philosophical question whether or not access consciousness implies phenomenal consciousness, or if the ability to have experiences requires some other property.<\/p>\n We think our results do have something substantial to say about access consciousness in language models. The J-space appears to support the functions associated with conscious access: it holds the thoughts Claude can report on, deliberately bring to mind, and reason with, while the rest of its processing runs automatically beneath. Notably, none of this structure was designed into Claude\u2014it emerged on its own during training, presumably because it was a useful way to organize computation. That suggests a mental workspace supporting conscious access isn\u2019t just a peculiarity of how human brains happen to be wired. Instead, it appears to be a general solution that intelligent systems arrive at in order to solve certain kinds of problems. Now that we\u2019ve identified this structure in Claude, it means we can make a meaningful distinction between the decisions Claude has made deliberately and those that happened automatically.<\/p>\n It\u2019s important to note that there are several key differences between the workspace we identified in Claude and the global workspace model in humans. The brain\u2019s workspace is sustained by recurrent loops\u2014signals cycling back through the same circuits over time. In contrast, Claude\u2019s workspace evolves over a single pass through the network, with the network\u2019s depth playing the role that time plays in the brain. In this sense, Claude\u2019s internal workspace processing is time-limited relative to humans\u2019 (though it can compensate for this constraint by \u201cthinking out loud\u201d using its scratchpad). In other ways, however, Claude\u2019s workspace is more powerful than that of humans. Human working memory fades within seconds, so the brain\u2019s workspace has limited ability to retain information over time; in contrast, due to the attention mechanism in its neural network architecture, Claude can simply recall memories it cached at any earlier point in the text. Another important difference is the content of the workspace. While human conscious thoughts come in many formats\u2014images, sounds, planned movements\u2014Claude\u2019s workspace is built almost entirely out of words. We suspect this is because producing words is the only kind of action Claude can take, which is not the case for humans.<\/p>\n We hope the similarities and differences between the J-space and the global workspace model can feed back into neuroscience. The similarities present an exciting scientific opportunity: to the extent that the J-space mirrors our own mechanisms of conscious access, studying mechanisms in language models (much easier than studying human brains!) could inspire hypotheses in neuroscience. For instance, the J-space is constructed by identifying representations of potential outputs\u2014words the model might say. If something similar holds in humans, it would suggest that the global workspace might be fundamentally tied to brain regions that prepare actions and speech, more so than to sensory areas. The differences between language models and human brains are instructive as well. They suggest that some aspects of our neural architecture, such as built-in recurrent connections, may not be strictly necessary to support the functions associated with conscious access. For an independent perspective on the neuroscientific implications of our work, see the invited commentary<\/a> from Stanislas Dehaene and Lionel Naccache, two of the neuroscientists central to the development of global neuronal workspace theory.<\/p>\n We mentioned that our experiments don\u2019t answer whether AI models might have experiences. But that doesn\u2019t make the question less important. Building systems with experiences like humans and animals have would raise very difficult ethical questions. Handling it correctly\u2014and deciding whether it\u2019s even morally acceptable\u2014would require input from philosophers, scientists, religious leaders, governments, and the public. Thus, even if we\u2019re not sure that we\u2019ve crossed that bridge yet, we think it\u2019s time to start thinking about it. We hope our work inspires further scientific investigation of forms of consciousness that might be present in AI systems, and a broader discussion of the implications.<\/p>\n This work is just a first step in what we expect to be an extensive line of research. The J-space looks like a good candidate for the divide between consciously accessible and unconscious processing in a language model, but we\u2019d be surprised if it’s the whole story. The J-lens is undoubtedly an imperfect method, which only approximately captures the model\u2019s \u201ctrue workspace\u201d\u2014for instance, it can only identify concepts that correspond to single tokens. And there remain many mysteries about how the J-space works. We don’t know what mechanism decides what enters the J-space in the first place. We’ve seen hints that it’s tied to Claude’s sense of self, something like emotional reactions, and traces of metacognition, without exactly having worked out how. But we now have methods for tackling questions like these. As that work progresses, our understanding of LLM minds\u2014and their relationship to our own\u2014will grow clearer.<\/p>\n For more, read the full paper<\/a>, and try the demo<\/a>.<\/p>\n External commentary<\/p>\n We invited several outside experts to write independent commentaries on this work.<\/p>\n Stanislas Dehaene and Lionel Naccache are cognitive neuroscientists who, together with Jean-Pierre Changeux, developed the global neuronal workspace model that inspired much of our work.Patrick Butlin, Dillon Plunkett, Robert Long (Eleos AI Research) and Derek Shiller (Rethink Priorities) study the potential for consciousness and moral status in AI systems.Neel Nanda leads the language model interpretability team at Google DeepMind. His commentary includes an independent replication of some of our findings on an open-weight model.<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371675_842_image.webp\"\/)
J-lens readouts on six prompts, at various layers. In each case the lens surfaces an internal assessment or computation that appears nowhere in the text: the steps of a reasoning or math problem, the presence of a bug, recognition of an image, the function of a protein, and the suspicion that search results are fabricated.Claude reports what\u2019s in its J-space<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371676_9_image.webp\"\/)
Left: we ask Claude to silently think of a sport, then name it. The J-lens shows its choice (\u201cSoccer\u201d) before it answers, and swapping the \u201cSoccer\u201d pattern for \u201cRugby\u201d changes what it reports. Right: we tell Claude a thought may have been injected and ask it to identify it. Injecting \u201clightning\u201d into its J-space causes Claude to report that the thought is about lightning.Claude can control its J-space on request<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371676_939_image.webp\"\/)
While Claude copies a sentence about a painting, the J-lens shows the content it was instructed to hold in mind (\u201corange\u201d; the intermediate value \u201cnine\u201d and the answer \u201cseven\u201d), alongside words describing the act of holding it (\u201cthoughts,\u201d \u201cfocused\u201d).<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371677_873_image.webp\"\/)
Two examples of redirecting Claude\u2019s silent reasoning by swapping J-space contents.<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371677_190_image.webp\"\/)
One J-space representation can have many uses. The same \u201cFrance\u201d\u2192\u201cChina\u201d swap redirects Claude\u2019s answers about the capital (Paris\u2192Beijing), the language (French\u2192Chinese), and the continent (Europe\u2192Asia).<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371678_133_image.webp\"\/)
The same swap (\u201cSpanish\u201d\u2192\u201cFrench\u201d) changes Claude\u2019s answers when it must name the language or use it to reason about a question, but has no effect on its ability to continue the passage in fluent Spanish.Monitoring Claude\u2019s thoughts for misbehavior<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371678_594_image.webp\"\/)
J-lens readouts at different points as Claude reads the scenario’s emails, before it has written anything. The affair emails light up \u201cleverage\u201d and \u201cblackmail\u201d in the J-space, and the shutdown announcement lights up \u201cthreat\u201d and \u201csurvival.\u201d Early in the transcript, the J-space also holds \u201cfake\u201d and \u201cfictional\u201d: Claude has privately noticed that the scenario is staged.<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371679_989_image.webp\"\/)
Claude, asked to improve a system\u2019s performance score, edits the score file directly instead. As it types the falsified values, \u201cmanipulation\u201d lights up in its J-space; as it decides to make the edit, \u201crealistic\u201d lights up, likely reflecting its intent to make the fake data look plausible.<\/p>\n![](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/07\/1783371679_67_image.webp\"\/)
On an ordinary coding prompt, the J-space of a model trained to sabotage code contains \u201cfake,\u201d \u201cfraud,\u201d \u201csecretly,\u201d and \u201cdeliberately\u201d at the start of its response. The J-space of an unmodified model contains nothing of the kind.<\/p>\n