AI Just SHOCKED Everyone: It’s Officially Self-Aware

Anthropic just showed that Claude can notice its own internal “thoughts.” Using concept injection, researchers perturbed Claude’s activations with vectors like “all caps” or “ocean,” and Claude 4/4.1 correctly flagged and named the injected concept at a sweet-spot strength—around 20% success with zero false positives on production models, peaking roughly two-thirds through the network’s layers. They also proved prefilling tricks can be made to look “intentional” by retroactively injecting evidence into prior activations—implying the model consults its previously computed intentions, not just surface text. This is early, unreliable, but real machine introspection.

Meanwhile, a University of Geneva & University of Bern study found six LLMs—ChatGPT-4, ChatGPT-o1, Gemini 1.5 Flash, Copilot 365, Claude 3.5 Haiku, DeepSeek V3—scored ~81–82% vs. 56% for humans on standardized emotional-intelligence tests, and even authored new, valid test items. Published in Communications Psychology.

???? Brand Deals & Partnerships: me@faiz.mov
✉ General Inquiries: airevolutionofficial@gmail.com

???? What You’ll See (sources):
• Anthropic research: Emergent introspective awareness in large language models
https://www.anthropic.com/research/introspection

• Technical explainer & examples (Anthropic/Transformer Circuits)
https://transformer-circuits.pub/2025/introspection/index.html

• Study on AI emotional intelligence
https://www.thebrighterside.news/post/artificial-intelligence-understands-feelings-better-than-people-study-finds/

• Paper: Large language models are proficient in solving and creating emotional intelligence tests (Communications Psychology, Nature)
https://www.nature.com/articles/s44271-025-00258-x

???? Why It Matters:
Models are starting to report on their own internal states and outperform humans on EI batteries. That boosts transparency and tooling for safer systems—while raising new questions about self-reports, intention, and control as models scale.