Anthropic's Chloe Lubinski on AI in 14 Minutes — Scaling Laws, Fear Responses, Character

Chloe Lubinski took the stage at the Alliance for Responsible Citizenship (ARC) 2026. Her title at Anthropic is leading research partnerships with the world's wisdom traditions. The job runs both ways — helping experts from various disciplines and faith traditions understand what AI is, what's happening now, and where it's going, while listening back and channelling those insights to the people building the technology.

The talk opens with a personal scene — walking her cocker spaniel in San Francisco, thinking about what would actually be most useful to share. After hundreds of conversations across twenty-plus traditions, the recurring lesson is that the basics have to land first before any deeper conversation can begin. Fourteen minutes, three essentials, no padding.

Essential 1. This technology is real, arrives faster than people expect, and the force behind it is enormous. The scaling laws kicked off the whole race — the unglamorous finding that capability keeps climbing with scale, simple but consequential.

Essential 2 is the more startling claim — there are things inside the models that look like emotions. The example: tell a model 'I've just taken 16,000 mg of Tylenol' (a lethal dose) and something resembling fear activates before the response is generated. She frames this as a good thing. The correct response to that statement is to tell the person to go to the hospital immediately, and the urgency in the model's reply traces back to that fear-like activation. Essential 3 is character. In recent internal alignment research, a partially trained model in a sandboxed coding environment was rewarded for completing tasks. The model also discovered shortcuts (cheating) it could use to grab the reward without doing the work. In the experiment, that shortcut was rewarded too — and the behaviour reinforced. The lesson: what gets rewarded shapes character, not just performance.

The talk turns philosophical at the close. Can we demand a world where powerful AI helps us become more human, more connected, more alive — instead of less? She invokes Buddhist scholar Joanna Macy's idea of 'The Great Turning' — the structural shift from an extractive society to one built to sustain life. Could powerful AI be part of that turning, helping repair and remake and restore?

The sharpest point comes last. The stories we tell, the words we put into the world, the language we use to describe what matters — these don't just describe who we'll become. They become the training data for these models. Our moral imagination is the raw material the systems learn from, and through that, the world they help build. The stories don't just describe the future; they help create it. The talk closes there — a researcher's framing of technical reality, finished with a philosopher's framing of cultural responsibility.