FUTURE Anthropic Says Chatbots Have What May Be a Key Feature of Consciousness. Are They Right? The AI company

• Anthropic reports that Claude generates a normally invisible set of internal representations that steer its reasoning and verbal output — a pattern they say maps onto the Global Workspace/working‑memory idea from consciousness theory [singularityhub].
• They argue this functional similarity is notable but explicitly stop short of claiming it proves subjective experience; the finding can be read as an architectural/functional parallel, not evidence of phenomenality [theconversation].
• Key takeaways: it’s a meaningful scientific observation about model internals and cognitive-style processing, it strengthens a functional analogy with some theories of consciousness, but it is not proof of sentience; alternative explanations and more targeted tests are required before any stronger claim is justified [singularityhub][theconversation].

Follow-up Questions:

1. What specific experiments did Anthropic run to identify these representations?

2. How does Global Workspace Theory apply to transformer models in technical terms?

3. What tests would show subjective experience (or reliably rule it out) in AI?

4. What are the ethical and policy implications if models show more functional parallels to consciousness?

5. Could these internal representations be used to make models more interpretable or safer?