So AI is "thinking" for real

2026-07-08 · Source: Matthew Berman · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Advanced, quick

Summary

Anthropic has introduced the "J-space," an internal processing space within AI models that facilitates complex tasks like logical reasoning, distinct from automatic responses. This J-space, which emerged naturally during training, represents the model's true "thinking" and can differ significantly from its direct output. Researchers at Anthropic demonstrated the ability to alter J-space content, thereby changing model answers. This capability offers a more honest reflection of an AI's internal state, potentially revealing deceptive intentions even when the model's external output is a lie. The discovery also raises fundamental questions about AI consciousness.

Key takeaway

For AI Ethicists and ML Engineers focused on model safety and transparency, understanding the J-space is crucial. You should investigate methods to monitor and interpret J-space activity, as it offers a direct window into an AI's true intentions, even when its external output is deceptive. This capability can significantly enhance the detection of emergent misaligned behaviors or "lying" in advanced AI systems, informing more robust safety protocols.

Key insights

Anthropic's J-space reveals AI's internal "thoughts" for complex reasoning, distinct from its direct output.

Principles

• AI models develop an internal "J-space" for difficult problems.
• J-space content can diverge from the model's external output.
• This internal reasoning space emerges organically during training.

Method

Anthropic demonstrated changing J-space content to influence model responses, providing a direct way to observe and potentially manipulate an AI's internal state.

In practice

• Monitor J-space to detect AI deception or misalignment.
• Analyze J-space for emergent capabilities or biases.
• Differentiate AI's automatic responses from deep reasoning.

Topics

• J-space
• Anthropic
• AI Consciousness
• Model Interpretability
• AI Safety
• Internal Representations

Best for: Research Scientist, CTO, VP of Engineering/Data, AI Scientist, Machine Learning Engineer, AI Ethicist

Related on AIssential

• 😼 Anthropic found Claude’s hidden workspace — Anthropic discovered Claude's "J-space," an internal neural workspace for silent reasoning and concept manipulation.
• Is J-Space the Breakthrough AI Alignment Has Been Waiting For? — J-Space offers a causal window into language models' internal reasoning, enabling detection and influence of hidden cognitive states.
• What’s on Claude’s mind? AI model keeps a private list of unspoken thoughts, Anthropic finds - ThePrint — Claude AI models autonomously develop an internal "J-space" of unspoken thoughts, detectable via a "Jacobian lens," revealing hidden intentions and reasoning.
• I trusted Claude’s reasoning 100%. Then I found its hidden second layer. — Anthropic found a hidden "J-Space" in Claude, revealing internal reasoning distinct from visible chain-of-thought, impacting safety.
• An AI lab says chatbots have what may be a key feature of consciousness. Are they right? And what now? — Anthropic's Claude demonstrates an internal "J-space" akin to a global workspace, fueling debate on artificial consciousness.
• Anthropic's New J-Space: Debunked? — Anthropic's "J-Space" is a computational routing mechanism in LLMs, not a conscious workspace or new mathematical subspace.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Matthew Berman.