# Will Douglas Heaven reports in MIT Technology Review: Anthropic finds a hidden space in Claude

[Canonical reading](https://escapemeridian.com/readings/will-douglas-heaven-reports-in-mit-technology-review-anthropic-1f421006)

MIT Technology Review’s newsletter, dated July 10, 2026, reports that Anthropic researchers used a tool called the Jacobian lens, or J-lens, to uncover a hidden area inside Claude that they named J-space. The newsletter says J-space contains words related to the response Claude is working on but may not ultimately produce. It compares these words to what might be “on its mind” before it speaks, while explicitly cautioning that Claude is not a person. The idea is that examining these words could offer a glimpse of a language model’s activity as it answers questions or carries out tasks. But the newsletter gives no examples of the words found, explains no method for interpreting them, and does not establish how reliably they relate to a model’s eventual response. It describes the findings as ranging from “the mundane to the unnerving,” then directs readers to the full story rather than detailing them here. The excerpt leaves unknown what Anthropic found in J-space and how those words related to Claude’s eventual responses.

Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.

## Original source

MIT Technology Review · Article

Original source date: Jul 10, 2026 (2026-07-10T12:10:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-19T16:43:09.967Z. First observed: 2026-11-19T11:48:04.611Z. Retrieved: 2026-11-19T11:50:00.686Z.

- [The Download: Claude’s inner workings and OpenAI’s “super app”](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Limits

Text as retrieved on November 19, 2026; originally published July 10, 2026. The newsletter points to a full story but does not provide its examples or explain how the J-lens works or how to interpret J-space. The excerpt does not establish how reliably the words in J-space relate to Claude’s eventual response.


## Open question

What did Anthropic find in J-space, and how did those words relate to Claude’s eventual responses?


## Source-stated

MIT Technology Review’s newsletter reports that Anthropic researchers built the Jacobian lens, or J-lens, and used it to identify a hidden area inside Claude called J-space.

> Researchers at the company built a tool called the Jacobian lens (or J-lens) and used it to uncover a hidden area, which they named the J-space, inside its flagship LLM, Claude.

- [Supporting source](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Source-stated

The newsletter says J-space contains words related to Claude’s response in progress that the model may not ultimately produce.

> The J-space contains words related to the response a model is working on but may not ultimately produce.

- [Supporting source](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Source-stated

The newsletter cautions that its comparison to what might be on Claude’s mind is figurative, not a claim that Claude is a person.

> If Claude were a person (which it is not), you might say these hidden words reveal what’s on its mind before it actually speaks.

- [Supporting source](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Source-stated

The newsletter characterizes Anthropic’s findings as ranging from mundane to unnerving, but does not detail them in this excerpt.

> What they found ranges from the mundane to the unnerving.

- [Supporting source](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Source and contributor context

MIT Technology Review · Publisher

The publisher is identified here. No individual author has been inferred.


## Source media

Media retrieval status: unchecked.


## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## Anthropic

Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems.

- [Explore this context](/entities/anthropic)
- [Reference](https://www.anthropic.com/news)
- [Reference](https://www.anthropic.com/research/team/economics)

## Claude

Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.

- [Explore this context](/entities/claude)
- [Reference](https://claude.com/product/overview)
- [Reference](https://www.anthropic.com/news)
- [Reference](https://www.anthropic.com/research/team/economics)

## Editorial provenance

Version 1. Publication ID: 172e2e41-0af5-c47c-da01-2ba5b9b6a49e.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-19T16:38:53.939Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical.

Observation: 9f800e5fadbf66c3508e164598b9a463ce37230cda0b98e0c27c76fc9e1aaeba. Source text hash: 73688b96c69f319a46f6c358ca2e98afe6a847f53aef210f0dd08bdf4fc7fa62.


## Revision and source-check history

### Version 1 — 2026-11-19T16:43:09.967Z

MIT Technology Review’s newsletter, dated July 10, 2026, reports that Anthropic researchers used a tool called the Jacobian lens, or J-lens, to uncover a hidden area inside Claude that they named J-space. The newsletter says J-space contains words related to the response Claude is working on but may not ultimately produce. It compares these words to what might be “on its mind” before it speaks, while explicitly cautioning that Claude is not a person. The idea is that examining these words could offer a glimpse of a language model’s activity as it answers questions or carries out tasks. But the newsletter gives no examples of the words found, explains no method for interpreting them, and does not establish how reliably they relate to a model’s eventual response. It describes the findings as ranging from “the mundane to the unnerving,” then directs readers to the full story rather than detailing them here. The excerpt leaves unknown what Anthropic found in J-space and how those words related to Claude’s eventual responses.

## Original source

MIT Technology Review · Article

Original source date: Jul 10, 2026 (2026-07-10T12:10:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-19T16:43:09.967Z. First observed: 2026-11-19T11:48:04.611Z. Retrieved: 2026-11-19T11:50:00.686Z.

- [The Download: Claude’s inner workings and OpenAI’s “super app”](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Limits

Text as retrieved on November 19, 2026; originally published July 10, 2026. The newsletter points to a full story but does not provide its examples or explain how the J-lens works or how to interpret J-space. The excerpt does not establish how reliably the words in J-space relate to Claude’s eventual response.


## Open question

What did Anthropic find in J-space, and how did those words relate to Claude’s eventual responses?


## Source-stated

MIT Technology Review’s newsletter reports that Anthropic researchers built the Jacobian lens, or J-lens, and used it to identify a hidden area inside Claude called J-space.

> Researchers at the company built a tool called the Jacobian lens (or J-lens) and used it to uncover a hidden area, which they named the J-space, inside its flagship LLM, Claude.

- [Supporting source](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Source-stated

The newsletter says J-space contains words related to Claude’s response in progress that the model may not ultimately produce.

> The J-space contains words related to the response a model is working on but may not ultimately produce.

- [Supporting source](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Source-stated

The newsletter cautions that its comparison to what might be on Claude’s mind is figurative, not a claim that Claude is a person.

> If Claude were a person (which it is not), you might say these hidden words reveal what’s on its mind before it actually speaks.

- [Supporting source](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Source-stated

The newsletter characterizes Anthropic’s findings as ranging from mundane to unnerving, but does not detail them in this excerpt.

> What they found ranges from the mundane to the unnerving.

- [Supporting source](https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/)

## Source and contributor context

MIT Technology Review · Publisher

The publisher is identified here. No individual author has been inferred.


## Source media

Media retrieval status: unchecked.


## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## Anthropic

Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems.

- [Explore this context](/entities/anthropic)
- [Reference](https://www.anthropic.com/news)
- [Reference](https://www.anthropic.com/research/team/economics)

## Claude

Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.

- [Explore this context](/entities/claude)
- [Reference](https://claude.com/product/overview)
- [Reference](https://www.anthropic.com/news)
- [Reference](https://www.anthropic.com/research/team/economics)

## Editorial provenance

Version 1. Publication ID: 172e2e41-0af5-c47c-da01-2ba5b9b6a49e.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-19T16:38:53.939Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical.

Observation: 9f800e5fadbf66c3508e164598b9a463ce37230cda0b98e0c27c76fc9e1aaeba. Source text hash: 73688b96c69f319a46f6c358ca2e98afe6a847f53aef210f0dd08bdf4fc7fa62.


### passed — 2026-11-19T16:38:53.939Z — anthropic/claude-opus-5.5

Publication: 172e2e41-0af5-c47c-da01-2ba5b9b6a49e

attribution — pass: This is one newsletter issue with commentary in the publisher's own words, not a bare listing. The lead item the reading uses is signed by Will Douglas Heaven and published by MIT Technology Review. The findings are correctly attributed to Anthropic researchers as reported by the newsletter.

> Read the full story on what they found.
—Will Douglas Heaven

summary — pass: The reading clearly states the actor (Anthropic), the tool (J-lens), the finding (J-space), and what J-space is said to contain. It keeps the source's figurative caveat and its 'mundane to the unnerving' framing. Gaps such as the missing examples and method are stated plainly. The question follows directly from what the excerpt leaves unknown. The other newsletter items are omitted, which is acceptable for a reading focused on the lead story.

> What they found ranges from the mundane to the unnerving.

date — pass: The July 10, 2026 date matches the UTC day of originalPublished (2026-07-10T12:10Z). The page is a publisher page, not an archive copy, so no archive disclosure is needed. The retrieval date is noted. The reading does not rely on later Deep Dive or related-story passages and uses no hindsight framing.

support — pass: Each claim matches a verbatim excerpt. The stated limitations (no examples, no method, unestablished reliability) are accurate descriptions of what the excerpt omits, not invented facts. No benefits or costs are asserted beyond the source.

> The J-space contains words related to the response a model is working on but may not ultimately produce.

media — pass: No media was selected, and no verified media candidates were supplied.

## Recorded references

Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter.

No public references recorded here yet.

[Reference data (JSON)](/readings/1f4210061c609432e65707575ea2842f339ab94a29fe3063d291f415e9e0f20f/references.json)

## Corrections and challenges

[Send a correction or an opposing source](https://escapemeridian.com/corrections/)

[JSON record](https://escapemeridian.com/readings/will-douglas-heaven-reports-in-mit-technology-review-anthropic-1f421006.json)
