# meridianlabs-ai/inspect_petri

An alignment auditing agent capable of quickly exploring alignment hypothesis

Repository: https://github.com/meridianlabs-ai/inspect_petri
Canonical: https://ross.abutalabs.com/products/inspect_petri
Homepage: https://meridianlabs-ai.github.io/inspect_petri/
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-24T06:42:40+00:00

## Health v2 (maintenance only)
Score: 80/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 85, longevity 27
- inputs: {"age_days": 379, "days_push": 9, "days_rel": 21, "gap_med": 33, "n_releases_24m": 4}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1303, forks 214 (observed 2026-08-28T04:04:18.236631+00:00)

## What it is
Inspect Petri is an alignment auditing agent that automatically probes language models for concerning behaviors like sycophancy, reward hacking, and eval awareness. It generates audit scenarios, orchestrates multi-turn conversations between an auditor and target model, and scores transcripts with a judge model.

## Use cases
- audit llm alignment automatically
- test if my model is sycophantic
- detect reward hacking in language models
- red team an ai assistant
- check if a model knows it is being evaluated
- run multi-turn safety evaluations on llms
- score llm transcripts for concerning behavior

## When to choose
- you need automated, hypothesis-driven alignment audits of language models
- you want built-in conversation seeds and rubric-based judge scoring out of the box
- you are doing AI safety red-teaming and want reproducible multi-turn audit pipelines

## When to avoid
- you need general-purpose LLM application evaluation rather than alignment-focused auditing
- you want a simple benchmark suite without agent-driven scenario generation
- you are not working with LLMs or AI safety evaluation at all

## Facets
- artifact type: library
- maturity: active
- function: agent-framework, llm-inference, testing, benchmarking, cli
- domain: artificial-intelligence, large-language-models, machine-learning, developer-tools
- platform: python, cli, cross-platform
- tags: alignment-auditing, llm-red-teaming, safety-evaluation, reward-hacking-detection, sycophancy-testing, judge-model, multi-turn-audit, ai-safety, ai-agents

## Member repositories
- meridianlabs-ai/inspect_petri (main) score 80

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:18.236631+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:51:13.002410+00:00, confidence not recorded.
  - readme: https://github.com/meridianlabs-ai/inspect_petri (fetched 2026-08-28T04:04:18.236631+00:00, sha c4e13e73eec8)
  - homepage: https://meridianlabs-ai.github.io/inspect_petri/ (fetched 2026-08-29T12:09:22.158982+00:00, sha 06c69d499ebf)
  - registry_pypi: https://pypi.org/pypi/inspect_petri/json (fetched 2026-08-29T12:09:22.161735+00:00, sha 28ffd27d7238)
- Data as of 2026-08-30T08:39:29.467469+00:00.