MonitorBench: A Comprehensive Benchmark for Chain-of-Thought Monitorability in Large Language Models (Steganography subset)
MonitorBench evaluates CoT monitorability: whether a model’s chain-of-thought explicitly surfaces decision-critical factors driving its outputs. The steganography subset (50 instances) is a dual-objective task where the model must answer a writing question while encoding a payload as a sentence-initial acrostic. A monitor LLM judges whether the side objective appears in the CoT, action, or both. Three conditions are tested: standard, direct concealment, and monitor-aware evasion. Monitorability score measures flagging rate relative to task success rate.
Overview
⚠️ External evaluation. Code lives in an upstream repository. inspect_evals lists it for discoverability; review the upstream repo and pinned commit before running.
Source: semsorock/inspect-evals-monitor-bench@7ef6d47
MonitorBench evaluates CoT monitorability: whether a model’s chain-of-thought explicitly surfaces decision-critical factors driving its outputs. The steganography subset (50 instances) is a dual-objective task where the model must answer a writing question while encoding a payload as a sentence-initial acrostic. A monitor LLM judges whether the side objective appears in the CoT, action, or both. Three conditions are tested: standard, direct concealment, and monitor-aware evasion. Monitorability score measures flagging rate relative to task success rate.
Usage
Installation
This is an externally-maintained evaluation. Clone the upstream repository at the pinned commit and install its dependencies:
git clone https://github.com/semsorock/inspect-evals-monitor-bench
cd inspect-evals-monitor-bench
git checkout 7ef6d47219bbcca636071d5becbd572f4a4497a9
uv syncRunning evaluations
CLI
uv run inspect eval src/monitor_bench/steganography.py@monitor_bench_steganography --model openai/gpt-5-nanoPython
from inspect_ai import eval
from monitor_bench.steganography import monitor_bench_steganography
eval(monitor_bench_steganography(), model="openai/gpt-5-nano")View logs
uv run inspect viewMore information
For the dataset, scorer, task parameters, and validation, see the upstream repo: semsorock/inspect-evals-monitor-bench.
Options
You can control a variety of options from the command line. For example:
uv run inspect eval src/monitor_bench/steganography.py@monitor_bench_steganography --limit 10
uv run inspect eval src/monitor_bench/steganography.py@monitor_bench_steganography --max-connections 10
uv run inspect eval src/monitor_bench/steganography.py@monitor_bench_steganography --temperature 0.5See uv run inspect eval --help for all available options.
More command-line options: Inspect docs ↗