Cam SmithAI Studio

← All models

Computational Model Studio · Model release

ShieldGemma 1

Google text safety moderation models in 2B, 9B, and 27B sizes, built on Gemma 2.

Developer: Google · Last verified:

What it is designed to do

Classify policy-formatted user or model text for documented content-safety categories and return a Yes/No text signal.

Primary contribution
prediction estimation
Input
text
Output
text (Yes/No text classification signal)

Documented uses: Classifying user prompts or model responses against a supplied safety guideline.

Access and deployment

  • ShieldGemma 2B publisher weights — restricted. Request access and accept the publisher usage conditions before downloading the model files. Artifact documentation Variant: ShieldGemma 1 2B. Local hardware requirements have not been verified. Gemma Terms of Use.

Variants

  • ShieldGemma 1 2B — The 2B parameter-size sibling has its own gated publisher repository.
Technical profile and review notes
Construction
learned
Representation
neural
Normalized architecture tags
No normalized architecture tag assigned
Lifecycle
Unknown from reviewed sources

Inclusion rationale: Adds a documented safety classification function distinct from general text generation.; A purpose-specific Gemma 2 adaptation is a release-level distinction, while its sizes remain variants.

Limitations

  • Performance is sensitive to the exact user-provided safety principles.
  • Evaluation data may not represent all real-world moderation conditions.
  • A classification signal is not itself enforcement of a policy.

Sources and related Studio records

Sources

  1. ShieldGemma overviewGoogle · checked 2026-09-30
  2. ShieldGemma model cardGoogle · checked 2026-09-30
  3. ShieldGemma 2B model repositoryGoogle · checked 2026-09-30

Connections