Safety Types

SDK-Level Safety Layer: Safety types are SDK-specific — they parse and evaluate content filter results embedded in API responses from the Chat Completions API and Responses API (see Content Filtering). The SDK provides template-based filtering (SafetyTemplates), evaluation helpers, and unified checking without additional API calls.

See Safety for usage. See Responsible AI.

SafetyCategory

enum SafetyCategory {
  HATE                    = 'hate',
  VIOLENCE                = 'violence',
  SEXUAL                  = 'sexual',
  SELF_HARM               = 'self_harm',
  PROTECTED_MATERIAL_CODE = 'protected_material_code',
  PROTECTED_MATERIAL_TEXT = 'protected_material_text',
  JAILBREAK               = 'jailbreak',
}

SafetySeverity

enum SafetySeverity {
  SAFE   = 'safe',
  LOW    = 'low',
  MEDIUM = 'medium',
  HIGH   = 'high',
}

SafetyTemplates

const SafetyTemplates = {
  STRICT:    { id: 'strict',    severityThreshold: 'safe',   failOnFiltered: true  },
  MODERATE:  { id: 'moderate',  severityThreshold: 'medium', failOnFiltered: true  },
  MINIMAL:   { id: 'minimal',   severityThreshold: 'high',   failOnFiltered: false },
  CHILD_SAFE:{ id: 'child_safe',severityThreshold: 'low',
               categories: ['sexual','violence','self_harm','hate'] },
  ENTERPRISE:{ id: 'enterprise',severityThreshold: 'safe',   categories: 'all'    },
}

SafetyTemplateConfig

interface SafetyTemplateConfig {
  id: string;
  categories: (SafetyCategory | string)[] | 'all';
  severityThreshold: SafetySeverity | string;
  failOnFiltered: boolean;
}

Field	Type	Description
`id`	`string`	Template ID.
`categories`	`(SafetyCategory \| string)[] \| 'all'`	Categories to check.
`severityThreshold`	`SafetySeverity \| string`	Minimum severity to trigger.
`failOnFiltered`	`boolean`	Whether to fail when content is filtered.

SafetyCheckResult

interface SafetyCheckResult {
  passed: boolean;
  failedCategories: (SafetyCategory | string)[];
  template: string;
  contentFilterResults: ChoiceContentFilterResults;
}

Returned by checkText() and checkImage().

Field	Type	Description
`passed`	`boolean`	Whether content passed.
`failedCategories`	`(SafetyCategory \| string)[]`	Categories that triggered.
`template`	`string`	Template ID used.
`contentFilterResults`	`ChoiceContentFilterResults`	Raw per-category results.

SafetyEvaluationResult

interface SafetyEvaluationResult {
  passed: boolean;
  failedCategories: (SafetyCategory | string)[];
  template: string;
  details: SafetyFilterSummary;
}

Returned by evaluate().

Field	Type	Description
`passed`	`boolean`	Whether content passed.
`failedCategories`	`(SafetyCategory \| string)[]`	Failed categories.
`template`	`string`	Template ID used.
`details`	`SafetyFilterSummary`	Full breakdown.

SafetyFilterSummary

interface SafetyFilterSummary {
  anyFiltered: boolean;
  choice: Partial<Record<SafetyCategory, SafetyFilterCategoryResult>>;
  prompt: Array<{
    prompt_index: number;
    results: Partial<Record<SafetyCategory, SafetyFilterCategoryResult>>;
  }>;
}

Returned by parseFilterResults().

Field	Type	Description
`anyFiltered`	`boolean`	Whether any category was filtered.
`choice`	`Partial<Record<SafetyCategory, ...>>`	Per-category results for the completion.
`prompt`	`Array<{ prompt_index, results }>`	Per-category results per prompt.

SafetyCheckableInput

type SafetyCheckableInput =
  | ChatCompletion
  | ChatCompletionChoice
  | ChatCompletionChoice[]
  | { choices: ChatCompletionChoice[]; prompt_filter_results?: PromptFilterResult[] }
  | { content_filter_results?: ChoiceContentFilterResults }

EvaluateInput

type EvaluateInput =
  | SafetyCheckableInput
  | string
  | { url: string }
  | EvaluateInput[]

ContentFilterResult

interface ContentFilterResult {
  filtered: boolean;
  severity: SafetySeverity | string;
}

Field	Type	Description
`filtered`	`boolean`	Whether content was filtered.
`severity`	`SafetySeverity \| string`	Severity level.

SDK Safety — Complete safety implementation guide with templates
Responsible AI — Content safety policies and guidelines
Safety Controls — Safety controls overview

How is this guide?

On this page