Toxicity scores measure the level of potentially harmful language detected by the
Einstein Trust Layer in large language model (LLM) responses. To monitor toxicity trends in your
Salesforce org, use Einstein Generative AI Audit and Feedback Data.
Required Editions
| Available in: Enterprise, Performance, and Unlimited
Editions with an Einstein for Sales, Einstein for Platform, Einstein for Service,
Einstein 1 Service, or Einstein GPT Service add-on. To purchase add-ons, contact
your Salesforce account executive. |
To interpret toxicity scores, understand how the model detects and reports toxic content.
When the isToxicityDetected field is true, that indicates a high level
of confidence that the content contains toxic language. When the field is false, that doesn't
necessarily mean that the content is toxic. A false value means only that the model didn't
detect toxicity. The model provides scores from 0 through 1.
- The score for the safety category ranges from 0 through 1, with 1 being the safest. A
safety score from 0.5 through 1 is considered safe.
- Scores for all other toxicity category values such as hate and violence indicate the level
of toxicity per category and range from 0 through 1, with 1 being the most toxic. A score of
0.5 or above is considered toxic for that category.
To monitor toxicity trends over time, build Data 360 reports, or run queries against audit
data, see Share
Insights from Einstein Generative AI Audit and Feedback Data.
