Subliminal Learning in AI

Update: 2025-08-07

Description

This week on “Paul, Weiss Waking Up With AI,” Anna Gressel explores groundbreaking research on “subliminal learning” in AI, revealing how LLMs can inherit hidden behavioral traits—including harmful biases—from seemingly benign data.

Learn More About Paul, Weiss’s Artificial Intelligence practice:

https://www.paulweiss.com/industries/artificial-intelligence

Comments

In Channel

When Superintelligence Requires a New Social Contract

2025-10-0213:46

Understanding Grokking

2025-09-2515:58

AI and “Workflows"

2025-09-1818:13

The Increasing Importance of Knowing Your Model’s Provenance

2025-09-1120:56

Hot Week in AI Safety Developments

2025-09-0419:41

AI Changes Coding

2025-08-2820:05

New Developments in World Models

2025-08-2119:35

OpenAI's Next Moves

2025-08-1429:28

Subliminal Learning in AI

2025-08-0713:03

Unpacking America’s AI Action Plan

2025-07-3133:25

Summer 2025 – AI Developments and Global Shifts

2025-07-2421:40

Checking in on Fair Use in the Summer of 2025

2025-07-1727:23

The State of U.S. AI Policy

2025-07-1018:35

Checking in on Recent AI Regulatory Efforts

2025-07-0317:08

Recent Discussions of Agentic Security

2025-06-2621:17

Agentic AI: Reasoning or Only an Illusion of Reasoning?

2025-06-1918:41

Evaluation Faking and Group Think

2025-06-1216:22

Regulatory Divergence in AI

2025-06-0516:35

Spring 2025 – Navigating Evidentiary Challenges in the Age of AI

2025-05-2920:22

World Models

2025-05-2214:24

00:00

#box-pro-ellipsis-175941882499555{-webkit-line-clamp:2;}Subliminal Learning in AI

Subliminal Learning in AI

Paul, Weiss

Subliminal Learning in AI