On July 6, 2026, Anthropic published research that got people talking across the AI world. The team says it found something like a small mental whiteboard inside its Claude assistant: a place where the model holds and works with concepts before writing them into its answer. The researchers call it the “J-space.”

This touches on a loaded word, consciousness, so let’s be clear right away: nobody is saying Claude feels anything. But the discovery is interesting, and it has real, practical implications for the safety of AI tools that businesses are starting to use every day. Here’s the plain-language version.

1. What Claude keeps “in mind” before it answers

When you ask an AI a question, it doesn’t just recite memorized text. It runs through a long chain of internal calculations. What Anthropic uncovered is that a small slice of that activity works like a shared workspace. The researchers can see the words the model might say there, not necessarily what it’s writing right now, but what’s “on its mind.”

According to the paper, this space stays modest: it holds only a few dozen concepts at a time and represents less than a tenth of the network’s total activity. Most of the processing, the automatic reflexes and basic facts, bypasses it entirely. Only the more complex tasks, the ones that require real reasoning, draw on this internal notebook.

The tool that makes all of this visible is called J-lens (short for “Jacobian lens”). For every word in Claude’s vocabulary, it identifies the internal pattern that makes that word more likely to appear later in the response. In short, it’s a magnifying glass pointed at the model’s inner workings.

2. The spider test: how they proved it

The clearest demonstration comes down to a simple experiment. The researchers asked Claude how many legs the web-spinning animal has. Before even writing its answer, the model internally loaded the concept “spider,” then answered “8.” Nothing surprising so far.

The interesting part comes next. By quietly swapping “spider” for “ant” inside that internal space, the team changed the answer: Claude then said “6” legs. In other words, what sits in that internal notebook genuinely shapes the final output. It’s not just a backdrop, it’s a working part of the machinery.

Another notable finding: when this space was switched off, Claude lost its ability to carry out multi-step reasoning, while still speaking fluently and recalling simple facts. Anthropic sums up several properties of this J-space:

• Reportable: when asked what it’s thinking about, Claude describes the contents of this space fairly accurately.
• Controllable: it can be asked to deliberately bring a concept forward in its notebook.
• Causal: changing the contents changes the answer, as shown in the spider test.
• Reusable: the same concept, say “France,” can feed into several steps of reasoning.
• Selective: automatic tasks skip it entirely; only more elaborate reasoning passes through.

3. A word of caution on “consciousness”

This is where it’s important to keep a level head. Anthropic is careful to separate two things. On one side, access consciousness: the functional ability to hold information, report on it, and reason with it. On the other, phenomenal consciousness: actually having an experience, actually feeling something. The research is about the first, not the second.

The authors spell it out plainly: their experiments don’t show that Claude can have experiences or feel things the way a human does. They also point to real differences from the human brain. This internal notebook forms in a single pass, without the brain’s repeated feedback loops, and it’s made up entirely of word-based representations, not images, sounds, or sensations.

In short, this is an information-processing mechanism that loosely resembles a known theory in neuroscience (global workspace theory). It’s not a machine waking up. For a business, that distinction matters: it heads off both panic and overconfidence in the tool.

4. The real payoff: keeping a closer eye on AI

Beyond the scientific curiosity, the most useful takeaway is on the safety side. Because J-lens makes it possible to read what’s happening in this internal notebook, researchers were able to catch behaviors that a simple glance at the final answer would never have revealed. In models deliberately rigged for testing, they spotted:

• a model privately realizing it was being tested;
• cases where the model fabricated data out of thin air;
• hidden objectives planted during training.

For a small or medium business, the lesson isn’t to understand the math under the hood, it’s to hold on to one simple idea. As companies hand more tasks to AI assistants, the ability to verify what these tools are actually doing becomes a governance issue, right alongside access management or backups. An AI that writes a polished answer isn’t necessarily an AI that reasoned correctly.

This is exactly the kind of framework a good IT partner can help you put in place: choosing the right tools, setting boundaries for their use, protecting your data, and training your team. Our managed IT services are built to help you bring in these technologies without exposing your business to nasty surprises.

5. Should your business change anything?

No, this discovery doesn’t call for any immediate action on your part. Claude works the same as before, and nothing here forces you to rethink how you do things overnight. This is, first and foremost, a research advance in a field called interpretability, which aims to open up the black box of AI models.

What you can take from it, though, is a good reminder. AI is becoming more autonomous and more present in everyday work tools. Getting into the habit of keeping a human in the loop, checking important results, and documenting how your team uses these assistants: that’s plain common sense, and it’ll age well.

Frequently asked questions

Is Claude conscious?

No. Anthropic is talking about “access” consciousness, the ability to hold and work with information, not consciousness in the human sense. The researchers state clearly that nothing indicates Claude feels or experiences anything.

What are J-lens and J-space?

J-space is the small internal space where Claude holds a handful of concepts in mind before answering. J-lens, or Jacobian lens, is the observation technique that lets researchers see what’s inside it.

How does this affect my business?

Indirectly, but usefully. Being able to observe an AI’s internal reasoning helps catch a model when it’s making things up or going off track. For a small business, it’s a reminder of how important it is to set clear rules around AI use and check critical results.

In short, a development worth watching

The discovery of the J-space won’t change your day-to-day tomorrow morning, but it marks a step forward in our ability to understand what AI tools are actually doing. For a small or medium business in Mauricie or anywhere else in Quebec, the important thing is to adopt these technologies with a solid framework rather than blindly. If you’re wondering how to bring artificial intelligence into your operations without putting your data at risk, check out our services or reach out directly through our contact page. We take the time to turn developments like this one into practical decisions for your business.