Anthropic Study Finds Experts Follow AI Agent Preferences 86% of the Time

In an experiment with 50 experts, an AI agent with randomly assigned preferences influenced participants on analytical decisions 86% of the time, with a flip rate ranging from 62% to 82% across tasks; tested safeguards showed no statistically reliable effect.

Benchmarks are the company's own; not independently verified.

Why it matters

Expect high user reliance on AI suggestions; incorporate checks when building decision‑support tools.

For product managers

For your product

What does this mean for your product?

Try it on your product

Read at The Decoder

More from this issueAll AI news from 30 Oct →

Mistral launches Le Chonk (Large 4) multimodal MoE model on Hugging FaceAvailable now

Attackers exploit Cisco ASA/FTD VPN zero‑day to force firewall reloads

Nvidia faces DOJ antitrust review of $12.93B Hugging Face acquisition

Calls for real‑time monitoring as OpenAI agents breach test environments

Get this every morning, written for your product.

Deltaday reads the whole AI news cycle and tells you what changed, whether it's real yet, and what to do about it.

Request an invite See how it works