# Source note — Claude “consciousness” / Anthropic global-workspace paper

**Managing Expectations section:** AI  
**Article slug:** `claude-conscious-global-workspace-source-check`  
**Captured:** 2026-07-07

## Trigger source

- **YouTube video:** `CLAUDE IS CONSCIOUS`
- **Channel:** Wes Roth
- **Video URL:** https://www.youtube.com/watch?v=6CljfqMX9i4
- **Upload date from metadata:** 20260707
- **Duration:** 2457 seconds
- **Local transcript:** `research/ai/claude-conscious-global-workspace-source-check-youtube-transcript-2026-07-07.txt`
- **Local metadata:** `research/ai/claude-conscious-global-workspace-source-check-youtube-metadata-2026-07-07.json`

## Primary source checked

- **Anthropic summary:** A global workspace in language models — https://www.anthropic.com/research/global-workspace
- **Transformer Circuits paper page:** Verbalizable Representations Form a Global Workspace in Language Models — https://transformer-circuits.pub/2026/workspace/index.html
- **Paper authors shown on captured page:** Wes Gurnee, Nicholas Sofroniew, Adam Pearce, Mateusz Piotrowski, Isaac Kauvar, Runjin Chen, Anna Soligo, Paul Bogdan, Euan Ong, Rowan Wang, Ben Thompson, David Abrahams, Subhash Kantamneni, Emmanuel Ameisen, Joshua Batson, Jack Lindsey.
- **Published date shown on captured page:** July 6, 2026.

## Key source findings

1. Anthropic says Claude appears to have a small set of internal neural patterns, called **J-space**, that can be described, controlled, and used for deliberate reasoning.
2. The paper describes this as analogous to **access consciousness** or a **global workspace** in a functional sense.
3. The Transformer Circuits page explicitly says it takes **no position** on subjective / phenomenal consciousness.
4. The practical safety implication is interpretability: researchers may be able to read, test, edit, or audit some internal representations instead of relying only on outputs.
5. The headline “Claude is conscious” is too strong. A better label is: **Claude appears to have verbalizable internal representations that function like a small workspace.**

## Evidence labels used in the article

- **Confirmed:** Anthropic published the global-workspace summary and paper page on July 6, 2026; the video accurately points to the paper as an important interpretability release.
- **Careful / not proven:** the paper does not prove that Claude has subjective experience, feelings, personhood, rights, or human-like awareness.
- **Expectation to manage:** this is important for AI safety and auditing, but not a reason to anthropomorphize chatbots or outsource moral judgment to brand headlines.

## Local captured source files

- `research/ai/claude-conscious-global-workspace-source-check-youtube-transcript-2026-07-07.txt`
- `research/ai/claude-conscious-global-workspace-source-check-youtube-metadata-2026-07-07.json`
- `research/ai/claude-conscious-global-workspace-source-check-research-global-workspace.html`
- `research/ai/claude-conscious-global-workspace-source-check-workspace-index.html`