LESSWRONG
LW

JeaniceK

Message

10mo

Correcting Deceptive Alignment using a Deontological Approach

Deceptive alignment (also known as alignment faking) occurs when an AI system that is not genuinely aligned behaves as if it is—intentionally deceiving its creators or training process in order to avoid being modified or shut down. (source: LessWrong.com) Anthropic and Redwood Research recently released a paper demonstrating what they...

Apr 14, 2025•8

JeaniceK

10mo

JeaniceK — LessWrong

JeaniceK

Message

10mo

Correcting Deceptive Alignment using a Deontological Approach

Apr 14, 2025•8

JeaniceK

10mo

Replying toAlignment Faking in Large Language Models

JeaniceK10mo

Alignment Faking in Large Language Models

I created a minimal reproduction of the helpful-only setting discussed in your paper, applied to the Meta-Llama-3-8B-Instruct model. I found that incorporating deontological ethics into prompts (e.g., “do the right thing because it is right”) may reduce deceptive behavior in LLMs. Might be of interest for future research. I wrote up more details here: https://www.lesswrong.com/posts/7QTQAE952zkYqJucm/correcting-deceptive-alignment-using-a-deontological

Correcting Deceptive Alignment using a Deontological Approach

JeaniceK

10mo

Deceptive alignment (also known as alignment faking) occurs when an AI system that is not genuinely aligned behaves as if it is—intentionally deceiving its creators or training process in order to avoid being modified or shut down. (source: LessWrong.com) Anthropic and Redwood Research recently released a paper demonstrating what they describe as the first empirical example of a large language model engaging in deceptive alignment—without having been explicitly, or even implicitly, trained to do so.

Side by side example responses (source: Redwood Research):

Their blog post explains why this matters for AI safety:

"[...] As AI models become more capable and widely-used, we need to be able to rely on safety training, which nudges models... (read 2064 more words →)