paper7Updated Nov 20, 2026 · Português
← All organizations

AI Safety Brazil

location not provided

4B models for prompt injection/jailbreak detection and moderation in pt-BR; 13 synthetic datasets and SafetyEval-BR leaderboard.

Models
2
Datasets
13
Downloads in 30 days
5.4K
Followers
2
Likes
3

Releases per quarter

Models and datasets, by creation date on Hugging Face.

2023202420252026

Model licenses

What each model card declares

  • cc-by-4.01
  • cc-by-sa-4.01

Dataset licenses

Commercial use under the declared license

  • Allowed11
  • With conditions1
  • Check1

Model types

By declared task

  • text-generation1
  • text-classification1

Where the models run

By file format and size

  • server2

Languages

Declared in model and dataset cards

  • pt15
  • en2

Models on Hugging Face (2)

Only what is published under the organization's Hugging Face account. Products and models under other accounts are not listed here.

ModelTaskRuns onSizeLicenseLikesUpdatedDownloads
guardbr-4btext-generationservercc-by-4.0021 days ago47
moderabr-4b-v3text-classificationservercc-by-sa-4.0021 days ago52

Datasets (13)

Commercial use reflects only the declared license; an open license does not guarantee that the content owners granted the rights.

DatasetLicenseCommercial useRowsUpdatedDownloads
moderabr-forensics-brcc-by-sa-4.0With conditions1K<n<10K21 days ago264
pix-scam-challenge-br-syntheticcc-by-4.0Allowed1K<n<10K22 days ago323
fato-ou-boato-brotherCheckn<1K23 days ago225
golpes-intencao-br-syntheticcc-by-4.0Allowed10K<n<100K23 days ago218
golpes-mensagens-br-syntheticcc-by-4.0Allowed10K<n<100K23 days ago291
jailbreak-br-syntheticcc-by-4.0Allowed10K<n<100K23 days ago895
moderacao-fronteiras-br-syntheticcc-by-4.0Allowed10K<n<100K23 days ago266
odio-contexto-br-syntheticcc-by-4.0Allowed10K<n<100K23 days ago291
overrefusal-br-syntheticcc-by-4.0Allowed10K<n<100K23 days ago269
pii-br-syntheticcc-by-4.0Allowed10K<n<100K23 days ago396
prompt-injection-br-syntheticcc-by-4.0Allowed10K<n<100K23 days ago857
safety-crosslingual-br-syntheticcc-by-4.0Allowed1K<n<10K23 days ago284
safetyeval-brcc-by-4.0Allowedn<1K23 days ago716

Declared links

Only what the authors declared in each model card. A missing declaration does not mean there is no use.

Trains on data from
the organization itself (jailbreak-br-synthetic, prompt-injection-br-synthetic, moderacao-fronteiras-br-synthetic and 1 more); dougtrajano (olid-br); mteb (told-br)
Builds on models from
Qwen (Qwen3.5-4B)