# Erik Hoel argues Anthropic’s global-workspace research faces measurement problems [Canonical reading](https://escapemeridian.com/readings/erik-hoel-argues-anthropic-s-global-workspace-research-faces-46e22aa2) Erik Hoel’s article was published July 13, 2026. Hoel argues that Anthropic’s research identifies reportable information in Claude, but does not establish that Claude is conscious. He focuses on J-space, which Anthropic describes as a way to identify internal representations available for verbal report. Hoel argues that measuring reportability may not independently test whether a system has a global workspace, a concept from consciousness research. Hoel also discusses an analysis of open-source models that, he says, does not show the sharp sensory/workspace/motor structure Anthropic reports for Claude Sonnet 4.5. He cautions that this is not an exact replication and says further work is needed. In Hoel’s view, the paper is useful for studying reportability and its possible role in multi-step reasoning, while illustrating broader difficulties in measuring consciousness. Whether the reported patterns distinguish a global workspace from other ways of producing reportable information remains an open question. 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 Erik Hoel · Article Original source date: Jul 13, 2026 (2026-07-13T16:34:01.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-10-29T17:57:28.035Z. First observed: 2026-10-23T21:49:54.737Z. Retrieved: 2026-10-23T21:51:40.141Z. - [Anthropic runs like Wile E. Coyote into the brick wall of consciousness research](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Limits Text as retrieved on October 23, 2026; originally published July 13, 2026. Hoel says the open-source model analysis is not an exact replication and that Anthropic has not released enough details to fully replicate its work. The article presents Hoel’s critique and interpretation; it does not independently establish whether Claude or other models are conscious. ## Open question What independent tests could distinguish a global workspace from other mechanisms that produce reportable information? ## Source-stated Erik Hoel’s article was published July 13, 2026. > Erik Hoel > Jul 13, 2026 - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Source-stated Hoel argues that Anthropic’s J-space analysis measures reportability and may not independently establish a global workspace or consciousness. > Conceptually then, this is like a mathematical measure of a disposition toward report; in other words, reportability. - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Source-stated Hoel says an analysis of open-source models does not show the sharp tripartite structure he discusses, but cautions that it is not an exact replication. > While it’s not an exact replication (which would be important to see—but Anthropic doesn’t actually release enough details to fully replicate anyway!) - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Source-stated Hoel describes the paper as concerning reportability and its possible usefulness for multi-step reasoning. > “Verbalizable Representations Form a Global Workspace in Language Models” is an interesting paper on the reportability of information in LLMs and how the property of reportability may be useful for multi-step reasoning in particular. - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Escape Meridian interpretation Escape Meridian reads Hoel’s central concern as whether the evidence distinguishes a global workspace from other mechanisms that support verbal report. > there are various conditions under which theories of consciousness become unscientific: e.g., they become unfalsifiable, or they become completely trivial, etc. - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Source and contributor context Erik Hoel · 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: 83291fa2-62a2-4ba5-9ed5-913304d46350. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T17:56:31.894Z. 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: 30186d72be8ce15738427f469c3c38d2e5e91a6ce615359394ae4a3069403910. Source text hash: f163e9edb4334929e8b7494c0b2e6811ab6707695c26d517b390b9ccfedeedcd. ## Revision and source-check history ### Version 1 — 2026-10-29T17:57:28.035Z Erik Hoel’s article was published July 13, 2026. Hoel argues that Anthropic’s research identifies reportable information in Claude, but does not establish that Claude is conscious. He focuses on J-space, which Anthropic describes as a way to identify internal representations available for verbal report. Hoel argues that measuring reportability may not independently test whether a system has a global workspace, a concept from consciousness research. Hoel also discusses an analysis of open-source models that, he says, does not show the sharp sensory/workspace/motor structure Anthropic reports for Claude Sonnet 4.5. He cautions that this is not an exact replication and says further work is needed. In Hoel’s view, the paper is useful for studying reportability and its possible role in multi-step reasoning, while illustrating broader difficulties in measuring consciousness. Whether the reported patterns distinguish a global workspace from other ways of producing reportable information remains an open question. ## Original source Erik Hoel · Article Original source date: Jul 13, 2026 (2026-07-13T16:34:01.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-10-29T17:57:28.035Z. First observed: 2026-10-23T21:49:54.737Z. Retrieved: 2026-10-23T21:51:40.141Z. - [Anthropic runs like Wile E. Coyote into the brick wall of consciousness research](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Limits Text as retrieved on October 23, 2026; originally published July 13, 2026. Hoel says the open-source model analysis is not an exact replication and that Anthropic has not released enough details to fully replicate its work. The article presents Hoel’s critique and interpretation; it does not independently establish whether Claude or other models are conscious. ## Open question What independent tests could distinguish a global workspace from other mechanisms that produce reportable information? ## Source-stated Erik Hoel’s article was published July 13, 2026. > Erik Hoel > Jul 13, 2026 - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Source-stated Hoel argues that Anthropic’s J-space analysis measures reportability and may not independently establish a global workspace or consciousness. > Conceptually then, this is like a mathematical measure of a disposition toward report; in other words, reportability. - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Source-stated Hoel says an analysis of open-source models does not show the sharp tripartite structure he discusses, but cautions that it is not an exact replication. > While it’s not an exact replication (which would be important to see—but Anthropic doesn’t actually release enough details to fully replicate anyway!) - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Source-stated Hoel describes the paper as concerning reportability and its possible usefulness for multi-step reasoning. > “Verbalizable Representations Form a Global Workspace in Language Models” is an interesting paper on the reportability of information in LLMs and how the property of reportability may be useful for multi-step reasoning in particular. - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Escape Meridian interpretation Escape Meridian reads Hoel’s central concern as whether the evidence distinguishes a global workspace from other mechanisms that support verbal report. > there are various conditions under which theories of consciousness become unscientific: e.g., they become unfalsifiable, or they become completely trivial, etc. - [Supporting source](https://www.theintrinsicperspective.com/p/anthropic-runs-like-wile-e-coyote) ## Source and contributor context Erik Hoel · 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: 83291fa2-62a2-4ba5-9ed5-913304d46350. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T17:56:31.894Z. 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: 30186d72be8ce15738427f469c3c38d2e5e91a6ce615359394ae4a3069403910. Source text hash: f163e9edb4334929e8b7494c0b2e6811ab6707695c26d517b390b9ccfedeedcd. ### passed — 2026-10-29T17:56:31.894Z — anthropic/claude-opus-5.5 Publication: 83291fa2-62a2-4ba5-9ed5-913304d46350 attribution — pass: The reading attributes the critique to Erik Hoel, the author and publisher of the piece. It attributes the J-space description and the paper to Anthropic, consistent with the source. > Anthropic runs like Wile E. Coyote into the brick wall of consciousness research Erik Hoel summary — pass: The summary captures Hoel's central argument: J-space measures reportability, which may not independently test for a global workspace. It also covers the open-source analysis with its caveats and his closing assessment. It defines J-space and labels global workspace as a consciousness-research concept, so a general reader can follow it. The question is a genuine open issue. > this paper does not prove much, other than how difficult measurement problems for consciousness are. date — pass: The stated date, July 13, 2026, matches the UTC calendar day of the original-publication instant and the article's dateline. Nothing is presented as occurring after that date. The source is a publisher page, not an archive copy, and the modification date equals the publication date. > Erik Hoel Jul 13, 2026 support — pass: Each claim is grounded in the source. These include Hoel calling J-space a measure of reportability, his caveat that the open-source analysis is not an exact replication, and his assessment of the paper's usefulness for multi-step reasoning. The Escape Meridian interpretation is labeled as such and is consistent with Hoel's unfalsifiability concern. The reading does not present the critique as independently verified. > While it’s not an exact replication (which would be important to see—but Anthropic doesn’t actually release enough details to fully replicate anyway!) media — pass: No media was selected, and no verified media candidates were supplied. There is therefore no media identity or relevance issue. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/46e22aa2366346da10fc53e30f74b237f49944b8c35754a7d087e7240b420d5b/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/erik-hoel-argues-anthropic-s-global-workspace-research-faces-46e22aa2.json)