Content Redaction for AI Gateway

Sensitive data stays out of AI — automatically

AI adoption in enterprises is accelerating, along with a growing issue: employees paste sensitive data (PII, financial records, credentials) directly into AI prompts. AI Gateway 1.6 addresses this with Content Redaction: Data Leak Protection (DLP) masking intercepts sensitive content before it reaches the AI model and again on the way out.
Your DLP policies now extend all the way into AI prompts and responses.
No new tools. No new policy engine. If you already have Netskope DLP, it now works on AI traffic — automatically.

What you can do with it:

Mask prompts	DLP policies detect and mask PII, credentials, and financial data in plain-text and JSON AI requests before they reach the model.
Mask responses	AI-generated responses are inspected and masked before delivery to the end user.
Plain text & JSON	Redaction works across plain-text prompts and structured JSON payloads.
Redaction event fields	New event fields show redaction status per transaction for audit visibility.

For more information, see Redact Sensitive Data.

UX & Configuration Boosts

Streaming Response Inspection

Your security checks no longer come at the cost of user experience

Previously, AI Gateway buffered the entire AI response before running content inspection, so users saw nothing until processing finished. For long responses like code generation or summaries, this caused noticeable delay. AI Gateway 1.6 introduces streaming inspection: content is inspected and forwarded as it arrives, preserving the real-time feel of AI interactions.

AI Gateway-CLI Diagnostics

Run appliance checks from the menu — no shell access needed

AI Gateway 1.6 adds a Diagnostics section to the AI Gateway – CLI interactive menu. Administrators can select and run diagnostic commands directly from the TUI — without unlocking the appliance or accessing the underlying shell. Every run is recorded in the appliance audit log.

Command	What It Tests
`ping`	Network reachability — tests connectivity to a target host
`traceroute`	Network path to a target host
`nslookup`	DNS resolution and hostname lookup
`nc`	Port connectivity — tests TCP/UDP connection to a host:port
`date`	System clock and time-zone correctness

Access: Available to all admins before and after enrollment. Each diagnostic run is logged to the appliance audit log.

Expanded AI Schemas & Providers

Three New Predefined AI Providers

More AI providers, same governance — no custom config needed

AI Gateway 1.6 adds three new predefined providers, giving your teams governed access to a broader set of AI platforms — without needing to configure custom providers manually. Select, apply policy, and go.

Custom Topics for AI Guardrails

Your guardrails, your rules — define content categories specific to your organisation.

AI Gateway 1.6 adds Custom Topics to AI guardrail profiles, letting you define organization-specific content categories for detection and policy enforcement.

Predefined categories don’t cover every organization’s risks. Custom Topics let you tailor content categories to your industry, policies, or risk areas without waiting for Netskope.

The updated LLM image remains backward compatible with AI Gateway 1.5 for predefined category detection; only custom topics require upgrading.

For more information, see AI Guardrails Policy.

AWS Bedrock Mantle Endpoint Support

Access more models via AWS Bedrock — AI Gateway now supports the Mantle endpoint

AWS Bedrock exposes two distinct endpoint types with different host URL formats. AI Gateway v1.6 adds full support for the bedrock-mantle endpoint, which previously could not be used with Claude models.

Endpoint Type	Host URL Pattern	Underlying Model	AI Schema	Example Provider
`bedrock-runtime`	`bedrock-runtime.{region}.amazonaws.com`	Any	AWS Bedrock	`cust-bedrock`
`bedrock-mantle`	`bedrock-mantle.{region}.api.aws`	Anthropic (Claude)	Claude	`cust-mantle-cla`
`bedrock-mantle`	`bedrock-mantle.{region}.api.aws`	OpenAI-compatible	OpenAI (Compatible)	`cust-mantle-ope`

Authentication: The authentication method for bedrock-mantle is identical to bedrock-runtime — no additional credential configuration is required.

For more information, see Configure AWS Bedrock as a Custom AI Provider.

Behind-the-Scenes Improvements

Custom Provider Hostnames Are Auto-Normalized

Envoy required lowercase. Now AI Gateway handles that for you.

Envoy, the underlying proxy in AI Gateway, requires strictly lowercase hostnames to route traffic correctly. Previously, configuring a custom provider with a mixed-case hostname could cause Envoy to fail routing silently — no error, just broken traffic. AI Gateway 1.6 automatically normalizes custom provider hostnames to lowercase before passing them to Envoy.

No action required. Normalization is applied automatically. Existing custom provider configurations with mixed-case hostnames will continue to work without any changes.

DLP Incidents Now Have Unique Titles

Goodbye, three identical incident titles

Incident objects now use the <transaction_id>/<activity>.json format, making each incident uniquely traceable. Previously, when multiple prompts were blocked by DLP in the same session, all incidents shared the same object name and title, making it impossible to match incidents to prompts.

ACTION REQUIRED

Enable Content Redaction
Content Redaction requires the DLP_ENABLE_REDACTION feature flag and must be activated.

Confirm the feature flag is active: GET /api/v2/policy/aig/dlp/featureflags.
Go to Policies → AI Gateway → DLP in your Netskope tenant.
Ensure your DLP profiles include the data identifiers to redact in AI traffic (PII, financial data, credentials).
Apply the policy to the relevant AI Gateway steering configuration.

Content Redaction does not apply retroactively to active sessions.

Review AI Gateway-CLI Diagnostics Access
The new Diagnostics section in AI Gateway – CLI is available to all admins without unlocking the appliance. Review CLI access before upgrading if you enforce strict least privilege policies.

Audit current AI Gateway – CLI admin access roles in your tenant.
Confirm Diagnostics access aligns with your support and operations team scope.

AI Gateway Release Notes Version - 1.6