richbc — LessWrong

From safety research prompt to cross-model universal jailbreak

This post describes a universal jailbreak discovery during work on black-box scheming monitors at MATS. The jailbreak itself is not released; see On publishing this post for details on infohazard considerations. This post is written in a personal capacity and all opinions contained here are my own, and not the...

Sep 328

Training Deliberative Monitors for Black-Box Scheming Detection

by aksh-n, adityasinha, Victor Gillioz, Simon Storf, Kilian Merkelbach, richbc, Axel Højmark, and Marius Hobbhahn

Paper: https://arxiv.org/abs/2605.29601 Thread: https://x.com/aksh_n0/status/2062568855814193497 TL;DR: Training small open-weight monitors provides a cost-effective alternative to prompted frontier monitors. Applying our training recipe to Qwen3.5-27B results in a monitor better at scheming detection than all smaller prompted monitors (Gemini 3.1 Flash-Lite, GPT-5.4 Nano, Claude Haiku 4.5) and Gemini 2.5 Pro, while achieving...

Jun 434

Constitutional Black-Box Monitoring for Scheming in LLM Agents

by Simon Storf, richbc, and Marius Hobbhahn

Paper: https://arxiv.org/abs/2603.00829 Thread: https://x.com/syghmon/status/2028878121051496674 Executive Summary Black-box monitors can detect scheming in AI agents using only externally observable information (no access to internal activations or chain-of-thought). We introduce two synthetic data pipelines, STRIDE and Gloom, for generating both benign and scheming trajectories. We optimize these monitors entirely on synthetic data...

Mar 328

richbc's Shortform

Dec 26, 20253

Base64Bench: How good are LLMs at base64, and why care about it?

This was a quick, short side-project produced during the MATS Research 8.1 extension. It's related to my group's main thread of work on black-box scheming monitoring through the connections to monitoring I explore below, but was time-boxed and pursued independently because I thought it was interesting! Executive Summary Figure 1....

Oct 5, 202539

On closed-door AI safety research

Epistemic status: Based on multiple accounts, I’m confident that frontier labs keep some safety research internal-only, but I’m much less confident on the reasons underlying this. Many benign explanations exist and may well suffice, but I wanted to explore other possible incentives and dynamics which may come into play at...

Aug 18, 202576

Building Black-box Scheming Monitors

by CorrigibleAgent, richbc, Simon Storf, and Marius Hobbhahn

Executive Summary * Our goal is to develop methods for training black box scheming monitors, and to evaluate their generalisation to out-of-distribution test sets. * We aim to emulate the real-world setting, where the training data is narrower and different from the data encountered during deployment. * We train our...

Jul 29, 202546