# safety-research/bloom

bloom - evaluate any behavior immediately  🌸🌱

Repository: https://github.com/safety-research/bloom
Canonical: https://ross.abutalabs.com/products/safety-research-bloom
Language: Python
License: MIT
License Family: permissive
Last push: 2026-05-07T20:47:35+00:00

## Health v2 (maintenance only)
Score: 55/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 81, release rhythm 35, longevity 31
- inputs: {"age_days": 435, "days_push": 118, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1392, forks 171 (observed 2026-08-28T04:04:36.143114+00:00)

## What it is
Bloom is a Python tool that automatically generates behavioral evaluation suites for LLMs, probing target models for behaviors like sycophancy, self-preservation, and political bias. It creates diverse test scenarios from a seed configuration, runs conversations with the target model, and scores the results.

## Use cases
- evaluate llm for sycophancy
- test if a model exhibits self-preservation behavior
- probe political bias in a chatbot
- generate custom behavioral benchmarks for a language model
- run automated safety evaluations on an llm api
- measure behavior stability under emotional pressure variations

## When to choose
- you need to probe a specific LLM behavior beyond fixed benchmarks
- you want reproducible, seed-configurable evaluation suites
- you're doing AI safety research on model behaviors

## When to avoid
- you need a fixed, widely-cited benchmark with stable results
- you want actively developed features - the repo is frozen and moved to Meridian Labs
- you need non-LLM software testing

## Facets
- artifact type: library
- maturity: maintenance
- function: llm-inference, testing, benchmarking, agent-framework
- domain: artificial-intelligence, large-language-models, machine-learning, developer-tools
- platform: python, cli, cross-platform
- tags: llm-evaluation, ai-safety, behavioral-testing, red-teaming, sycophancy, benchmark-generation

## Member repositories
- safety-research/bloom (main) score 55

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:36.143114+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 2026-08-30T04:39:33.366167+00:00, confidence not recorded.
  - readme: https://github.com/safety-research/bloom (fetched 2026-08-28T04:04:36.143114+00:00, sha d7569d2e013d)
- Data as of 2026-08-30T08:39:29.467469+00:00.