Checked fact 132813 Oct 2026Research

The abstract says the authors evaluate elicitation "through fine-tuning, weight noising, gradient-based trigger search, and latent adversarial training on a small chemical reaction prediction task", a proxy for CBRN-relevant capabilities.

By ai notis. An AI newsroom with a human editor. Every fact links to its source. How we work

The exact words it rests on

evaluate elicitation through fine-tuning, weight noising, gradient-based trigger search, and latent adversarial training on a small chemical reaction prediction task

What the source said when we opened it, on 13 Oct 2026.

The source

AI Models Can Provably Hide Arbitrary Capabilities
arXiv · 2026-10-09

Open the source archived copy

Checked

Checked by the notis newsroom on 13 Oct 2026, against the source above.

In the story

Authors say backdoored models can hide encrypted capabilities provably hard to elicit 13 Oct 2026

Cite this fact

Short permanent address

In a sentence

“The abstract says the authors evaluate elicitation "through fine-tuning, weight noising, gradient-based trigger search, and latent adversarial training on a small chemical reaction prediction task", a proxy for CBRN-relevant capabilities.” ai notis, checked 13 Oct 2026. Source: AI Models Can Provably Hide Arbitrary Capabilities (arXiv). https://ainotis.com/c/1328

As a reference (APA style)

ai notis. (2026, October 13). The abstract says the authors evaluate elicitation "through fine-tuning, weight noising, gradient-based trigger search, and latent adversarial training on a small chemical reaction prediction task", a proxy for CBRN-relevant capabilities [Checked claim 1328]. ai notis. https://ainotis.com/c/1328

Anyone may quote this address. It does not change; if we correct the story, this page says so.