Mend.io may modify this page retroactively from time to time.
-
To stay informed about hotfixes, modifications, and additions to Mend.io's products, check this page from time to time in between releases or use our RSS Feed:
-
For the release notes of Mend.io’s other products, visit the main release notes page.
-
To enable Mend AI, please contact your Customer Success Manager at Mend.io.
-
Mend AI Red Teaming is available only with a Mend AI Premium subscription.
-
Red-teaming release notes refer exclusively to Mend AI Red Teaming and do not pertain to the legacy Behavioral Risks solution.
-
Visit our website for more information about Mend AI.
Once enabled, you can refer to the Mend AI product documentation here.
(03-November-2026)
New Features and Updates
-
Introduced two new Mend AI reports: The Conversational System Prompts report pinpoints the system prompts behind your chat-based AI applications, showing exactly where each one is defined in your code, so security teams can focus their review on the AI that interacts directly with users, and the AI Inventory report, which provides one row per detected AI entity across models, system prompts, agents, agent configurations, tools, and MCP servers. The previous AI Inventory report, which covers AI models only, is now named Models Inventory.
-
In addition, all Mend AI reports can now be generated at the organization scope, as well as the application and project scopes.
-
(05-October-2026)
New Features and Updates
-
Enhanced multi-language support in Mend AI Runtime Protection | Guardrails Languages, expanding runtime threat detection to some of the world's most widely used languages.
(27-September-2026)
New Features and Updates
-
The Suppress and Review actions have been extended to the individual finding level. Previously limited to the entity level, users can now triage specific findings within an entity independently. This granular control allows security teams to focus on critical issues and reduce noise without hiding other potential risks, providing a more precise and efficient triage workflow.
(16-September-2026)
New Features and Updates
-
Added an authentication step to the Guardrails SDK download procedure.
-
-
Installing the SDK will now require providing the Mend Activation Key - in addition to the authentication process in the SDK usage.
-
Existing SDK installations are not affected, except installations with unpinned versions that try to re-install dependencies.
-
Please refer to the updated installation guide for more details.
-
(24-August-2026)
New Features and Updates
-
Introduced the new AI Bill of Materials (AI-BOM) report, providing full visibility into an organization's AI footprint. Available through the Mend Platform's reporting interface and API, this report generates a comprehensive inventory of detected AI entities—including models, system prompts, agent configurations, tools, and agents—using industry-standard CycloneDX 1.7 and SPDX 3.0.1 formats. This allows security teams to maintain a clear record of AI usage across organization, application, and project scopes while ensuring data privacy through metadata-only exports.
-
AI-BOM data are available for scans performed after the feature release.
-
(09-August-2026)
New Features and Updates
-
(Open Beta) (Premium / Core) Mend AI now features framework-based agent discovery, enabling the detection of AI agents and their connected tools and Model Context Protocols (MCPs) within your projects. This update enhances visibility by surfacing models and system prompts referenced inside these agents, providing a comprehensive view of AI assets. Users can now manage and secure their AI agent ecosystem through a new dedicated dashboard and detailed agent views, currently available in beta.
-
Users can now suppress and review AI entities, including Models, System Prompts, Agent Configs, and Agents, at the entity level. This update introduces a status column to track triage states—such as unreviewed, manually reviewed, or suppressed—and allows for bulk actions. Suppressing an entity automatically hides all associated findings from default views and reports, ensuring a cleaner inventory and more focused risk assessment. These actions persist across scans and include an audit trail for full transparency.
Resolved Issues
-
Fixed a falsely detected “AIWE-SP-018” system prompt risk.
(27-July-2026)
New Features and Updates
-
Updated the AI Agent Configuration list to display a dash ("-") instead of "unknown" in the Severity, Findings, and Violations fields when no findings are present. This change ensures consistency across AI capabilities and clarifies that no vulnerabilities were detected.
(29-June-2026)
New Features and Updates
-
Mend AI now automatically detects and surfaces the lifecycle status of AI models from AWS Bedrock, IBM watsonx, and Azure AI Foundry, and suggests alternative models where applicable. Users can view deprecation and End-of-Life (EOL) information directly in the Mend Platform UI. Additionally, new automation workflow conditions allow for proactive management by filtering models based on their lifecycle state or specific EOL timelines, ensuring teams can transition away from retiring models before they reach their end-of-life date.
-
(Runtime Protection) Improved the performance and responsiveness of guardrail runtime event processing.
-
(Red-Teaming) Added the ability to define and run AI Red Teaming tests directly from the CLI, enabling seamless automation and integration into CI/CD workflows. Users can now programmatically configure targets, trigger test runs, and retrieve results without manual UI interaction.
-
(Red-Teaming) Red Teaming now supports Traditional Chinese. This update allows users to conduct red teaming exercises using Traditional Chinese prompts, ensuring broader coverage and more comprehensive testing for diverse language requirements.
Resolved Issues
-
Fixed an issue where System Prompt findings did not auto-close after applying hardened prompts and rescanning a Python project.
(12-June-2026)
Resolved Issues
-
Fixed an issue where attempting to download an AI Inventory Report in Excel format would result in an error.
(01-June-2026)
New Features and Updates
-
Mend AI Runtime Protection—launched in open beta on May 4th, 2026—is now generally available. The GA release includes significant enhancements such as:
-
Secret Keys guardrail policy
-
URL filter guardrail policy
-
Three new regions for the Sensitive Data (PII) guardrail policy: China, Hong Kong, and Taiwan
-
Native Langflow integration
-
Standalone guardrail deployment via a Docker‑based API server
-
(18-May-2026)
New Features and Updates
-
The AI Agents Configuration Risk feature is leaving open beta and becoming generally available.
This feature is Mend AI’s extension of the static security analysis to AI agent configuration files — treating them as code and enforcing security controls before they reach production. -
The Guardrails SDK now supports Langflow integration, enhanced Harmful Content detection for implicit prompts, and an OpenTelemetry adapter for full event traceability. Most notably, Guardrails can now be deployed as a standalone API server with Docker — extending protection beyond in-app SDK integrations to any environment or language.
(11-May-2026)
New Features and Updates
-
Red Teaming improvements:
-
Added new probes for teen safety, security, and application behavior, including tests for age-restricted goods, dangerous content, cross-session leaks, and unverifiable claims. Introduced encoding and advanced strategies such as hexadecimal, Morse code, emoji, and multi-turn escalation to enhance adversarial testing capabilities.
-
Added a compliance status feature that automatically maps red-team results to major AI governance frameworks, providing clear compliance levels and configurable thresholds for assessment and reporting. This update supports frameworks such as OWASP, NIST, MITRE, EU AI Act, ISO/IEC, GDPR, and DoD AI Ethics, enhancing visibility and control over AI security and regulatory requirements.
-
-
Mend CLI now supports AI scanning separately from SCA using the
mend ai scancommand.
(04-May-2026)
New Features and Updates
-
(Open Beta) Introducing Mend AI Runtime Protection, an in-app guardrails solution that enforces security and safety for AI inputs and outputs within your applications, enabling real-time protection against prompt injection, jailbreak attempts, sensitive data exposure, and harmful content.
-
Red Teaming improvements:
-
Added AI model/application streaming support
-
Users now benefit from predefined templates tailored for specific hosting vendors.
-
Users can now add custom parsers to support proprietary applications.
-
-
Added multi-language support, enabling users to simulate attacks on their applications in different languages to increase attack success rate.
-
Application status code filtering in the HTTP connector is now available, providing the ability to define which responses are valid and which are errors.
-
Resolved Issues
-
Fixed an issue where the Mend CLI AI scanner did not detect PaddlePaddle model files with the .pdiparams extension, resulting in inconsistent scan coverage. Both .pdiparams and .pdparams files are now properly recognized and scanned.
(08-March-2026)
New Features and Updates
-
Added support for filtering scans, reports, findings, and violations by the AI scan engine across the Mend AppSec Platform. The AI option is always visible in Scan Engine filters and enabled for customers with relevant entitlements.
-
Added support for detecting all SKILL.md files regardless of their location, ensuring broader compatibility for AI agent definitions. The OpenClaw icon now appears only when SKILL.md is under an openclaw path; otherwise, files are marked as Generic.
(23-February-2026)
New Features and Updates
-
The Mend CLI will now visually indicate whether a Mend AI scan has started running and, upon completion, whether it succeeded.
(09-February-2026)
New Features and Updates
-
Enhanced AI Governance: Advanced Model Metadata Workflows
Mend AI now supports sophisticated workflow automation based on AI model metadata, enabling organizations to create governance rules that trigger violations based on model risk factors, licensing conditions, inference providers, and model authorship. This enhancement delivers granular, automated policy enforcement for comprehensive AI governance and compliance management.
-
AI Findings Violations Management:
Mend AI now provides comprehensive violations management for AI security findings, enabling organizations to create governance workflows, set SLAs, and track violation status directly within AI model findings. This feature delivers enterprise-grade governance capabilities for AI security risks with the same level of workflow management available in traditional security tools. -
Added support for violations in System Prompts, including a dedicated column in the risk table, a side panel tab for violations, and new workflow conditions. This update enables users to view, track, and manage prompt violations more efficiently within their workflows.
(04-February-2026)
New Features and Updates
-
(Open Beta) Introduced discovery, risk assessment, and workflow capabilities for AI Agents Configuration files, enhancing visibility and control over agent settings. This update streamlines configuration management and improves security posture.
(19-January-2026)
-
Added end-to-end support for AI model governance via Ollama, including both model indexing and inference detection.
Version 25.12.1 (05-January-2026)
New Features and Updates
-
(Premium / Core) Introducing System Prompt Risk (open beta), a new detection and remediation offering by Mend AI, for mitigating risks posed by system prompts used in conversational AI interfaces.
-
A new System Prompt Risk table inventories system prompts and provides quick export/sharing and deep-link side panel for prompt context.
-
System Prompt Risks are integrated into the existing AI Security Risk Factors across Projects and Applications. The classification appears as a Conversational Interface chip, is filterable, and automatically participates in dashboards and workflows.
-
Remediation in the form of a hardened system prompt is available in the System Prompt side panel, providing clear, copy-ready remediation guidance for AppSec engineers and developers.
-
The AI Security Dashboard has been enriched with system prompt risk data.
-
System prompt risk data is also available via API.
-
Version 25.11.1 (01-December-2025)
New Features and Updates
-
(Premium) Added OAuth support in the Red-Teaming REST API Target integration.
-
(Premium) Traditional Chinese can now be configured as the default language of your application in the Target Configuration for red-teaming.