Customer-Facing Release Notes
AI Discovery (Cloud)
- Google Vertex AI model scanning — You can now scan and track Google Vertex AI models as part of your AI resource inventory, expanding cloud AI discovery to cover Vertex AI deployments.
- Anthropic integration for cloud scanning — You can now connect your Anthropic account to monitor and scan Anthropic API usage across your cloud infrastructure.
- Improved handling of Google Vertex AI permission errors — Google Vertex AI integrations now automatically disable and notify you when required permissions are missing, instead of retrying indefinitely without explanation.
- Duplicate software entries prevented — AI software discovery no longer creates duplicate entries when scans run multiple times, ensuring your inventory stays clean and accurate.
AI Discovery (Workforce)
- Anthropic Claude Code workforce user discovery — You can now discover and track Anthropic Claude Code users across your workforce through a new Anthropic integration.
- Discovered Topics for AI workforce activity — A new Discovered Topics tab shows you topics automatically identified from your workforce's AI usage, helping you understand what subjects employees are discussing with AI tools.
- Expanded AI application detection — Microsoft 365 and Google Workspace scanners now detect additional AI platforms including DeepL, Ideogram, and Poe.
- Microsoft 365 authentication fix — Fixed an issue where the Microsoft 365 OAuth sign-in popup could fail due to browser security policies, ensuring reliable connection setup.
- Google Workspace credential handling — Google Workspace integrations now automatically disable and notify you when OAuth credentials expire, instead of failing silently or crashing.
AI Logging (Cloud)
- Azure OpenAI log collection — You can now collect and monitor chat completion logs from your Azure OpenAI deployments, giving you visibility into AI usage across your Azure infrastructure.
- Amazon Bedrock response capture — The endpoint agent now fully captures and decodes Amazon Bedrock AI responses across all supported formats, enabling policy enforcement and logging for Bedrock-based AI usage.
- Token usage estimation for all logs — When an AI provider does not report token counts, FireTail now automatically estimates them, ensuring you always have reliable token usage metrics across all AI logs.
- Improved log processing performance — AI log processing is now significantly faster thanks to optimized parsing and pre-loaded models, resulting in quicker availability of log data.
AI Logging (Workforce)
- Guardrail evaluation details in logs — Workforce logs now include detailed information about which guardrails were evaluated, whether they blocked requests, and what interventions were triggered.
- Opera browser support — The browser extension now supports Opera, expanding workforce AI monitoring to users of this browser.
- Mistral AI monitoring — You can now monitor Mistral AI chat streaming activity through the browser extension.
- Cohere and Perplexity AI support — The endpoint agent now captures and monitors responses from Cohere and Perplexity AI services.
- AI response capture for streaming services — The endpoint agent now captures AI responses from streaming connections for ChatGPT, Claude, Gemini, OpenAI, and Microsoft Copilot, enabling complete logging of AI interactions.
- Improved markdown rendering in log views — AI log messages and employee log messages now render in markdown format, making conversations and responses easier to read.
- WebSocket-based ChatGPT monitoring fixed — Fixed response capture for WebSocket connections, resolving issues with monitoring ChatGPT activity.
- Reduced duplicate logging — Fixed issues where repeated visits to the same AI service or file upload actions generated redundant log entries, reducing noise in your log data.
AI Governance
- Topic guardrail management — You can now create, edit, and manage topic-based guardrails for AI Workforce Policies, giving you fine-grained control over which topics are allowed or blocked in AI interactions.
- Local guardrail evaluation with multilingual support — The endpoint agent now evaluates guardrails locally for faster enforcement, with multilingual support including non-English prompts that automatically fall back to remote evaluation.
- Dedicated guardrail block page — When a guardrail blocks an AI interaction, users now see a clear block page in their browser explaining which guardrail was triggered, with deduplication to prevent repeated notifications.
- Improved guardrail reliability — Guardrails that are unavailable or still being prepared no longer cause request failures. They are gracefully skipped and reported separately, ensuring policy enforcement remains reliable.
- Asynchronous guardrail topic processing — Guardrail topics are now automatically processed in the background to generate example prompts and embeddings, enabling more accurate content enforcement without manual setup.
- Real-time policy updates via WebSocket — The endpoint agent now receives policy and rule updates in real time, so changes you make to governance policies take effect immediately without waiting for scheduled refreshes.
- Topics management — A new Topics page lets you view and manage custom, system, and suggested topic guardrails for controlling AI workforce interactions.
- International character support — Names, descriptions, and input fields across the platform now support accented Latin characters, Japanese scripts, and other Unicode characters.
- High-volume policy notifications — Resource policies that match large numbers of items now correctly count and report all matches in notifications, respecting your configured notification thresholds.
- Configurable Microsoft Teams bypass — The endpoint agent now includes a configurable proxy bypass for Microsoft Teams and Microsoft 365 domains, enabled by default, to prevent the agent from interfering with these tools.
- Reduced CAPTCHA triggers — The endpoint agent now uses techniques to reduce CAPTCHA and bot detection triggers during normal browsing while AI monitoring is active.
Posture Management
- AI Risk Dashboard — A new AI Risk Dashboard provides a consolidated view of AI-related risks across your organization.
General Improvements
- Simplified navigation for single-product organizations — Organizations using a single FireTail product now see a streamlined navigation with simplified tabbed pages for employee, inventory, policy management, and settings.
- Reorganized navigation structure — AI-related routes have been moved up one level in the navigation for easier access, and Model Scanning and Alerting are now found under Workload paths.
- Workforce Usage and Risk dashboards — Workforce dashboards have been reorganized into separate Usage and Risk views, reducing redundancy and making it easier to find the information you need.
- Organization milestones — A new progress indicator in the navigation bar shows your organization's setup completion status, helping you track onboarding steps.
- Dashboard navigation redesign — Dashboard navigation now uses a carousel with clickable indicators instead of dropdown menus.
- Simplified agent setup — The endpoint agent and browser extension now automatically detect your environment from the project token, eliminating the need to manually configure region and platform settings.
- SHA256 checksums for endpoint agent — SHA256 checksum files are now published alongside all endpoint agent installers, allowing you to verify the integrity of downloaded binaries.
- Improved empty states — Resource listings across the platform now display informative guidance messages when empty, tailored to your permissions.
- Browser extension reliability improvements — Configuration and policy updates in the browser extension now persist across browser restarts and service worker terminations, ensuring continuous protection.
- Self-serve organization creation removed — You can no longer create organizations directly from the UI. To create a new organization, contact support.
- Improved service reliability — Fixed several minor bugs and stability improvements across the platform, including better error handling, retry logic, and connection management.
