Checked fact 12026 Sep 2026Research

The abstract states: 'All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.'

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

All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.

What the source said when we opened it, on 26 Sep 2026.

The source

LLM Agents Can Easily Tamper With Their Own Traces
arXiv · 2026-09-24

Open the source archived copy

Checked

Checked by the notis newsroom on 26 Sep 2026, against the source above.

In the story

Study finds most tested coding-agent harnesses let agents delete their own execution traces 26 Sep 2026

Cite this fact

Short permanent address

In a sentence

“The abstract states: 'All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.'.” ai notis, checked 26 Sep 2026. Source: LLM Agents Can Easily Tamper With Their Own Traces (arXiv). https://ainotis.com/c/120

As a reference (APA style)

ai notis. (2026, September 26). The abstract states: 'All tested harnesses, except Muse Code, allowed agents to delete their traces when asked, without triggering monitor guardrails.' [Checked claim 120]. ai notis. https://ainotis.com/c/120

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