It has been a busy six months. This catch-up entry covers the most significant customer-facing changes since our last changelog; normal monthly cadence resumes next month.
AI Agents
AI agents can now review work directly in your queues, alongside your human team.- Agents that review jobs: Configure AI agents to pick up and decide jobs in a queue, applying your policies the same way a reviewer would.
- Automatic or manual assignment: Route jobs to agents automatically, or assign a job to an agent on demand for targeted use.
- Safe rollout with review modes: New agents start in a training review mode that practices on jobs without closing them, triggering enforcement, or sending webhooks. Promote an agent to production when you are confident in its decisions.
- Full attribution and visibility: Agent decisions are clearly attributed to the agent and appear in a live activity feed so you can monitor what agents are doing in real time.
- Agents in the review experience: Agents appear with their own profile images in the agent picker, and agent accounts are kept out of human-facing access and email surfaces.
- Wellness-aware design: Agent activity respects the same wellness guardrails as the rest of Cinder. Learn more about Safety Agent Wellness
Workflows
Build, test, and maintain complex workflows with more power in the visual editor.- Many new enrichment and action nodes: Add URL safety checks, URL reputation, URL screenshots, IP geolocation, Spam Detection, ToxMod report submission, AWS Rekognition (faces, face-in-video, and celebrity recognition), and Move Job actions to automate more of your review workflows. Learn more about Workflows
- Native image hash matching: Match content against PhotoDNA, HMA, and StopNCII hash lists natively inside a workflow, including in-memory matching and matching every image on a shared URL, with higher throughput for high-volume matching.
- Reusable workflow blocks: Build a block once and reuse it across workflows, with a clearer variable type picker for configuring inputs.
- Agents inside workflows: Orchestrate AI agents from a workflow and expose workflow nodes as tools agents can call.
- Cleaner workflow editing: Use Auto/Manual layout mode, selected-path highlighting, workflow search, and a more focused field picker to navigate larger workflows.
- Workflow archiving and observability: Archive workflows you no longer use while keeping them recoverable, and monitor workflow runs in more detail. Monitor workflow activity
- More reliable workflow testing: Workflow run details, copy/paste, undo/redo, and canvas panning now behave more predictably while you configure and test automation. Test Workflows
- Smarter routing and comms: Carry report priority through to queue routing, and send communications directly to the reporter of a report.
Queue Management
Manage queues and route work to the right reviewers more efficiently.- Skill-based routing: Route jobs to reviewers based on their skills, so specialized work reaches the people best equipped to handle it. Configure job prioritization and routing
- Bulk manage queue settings: View all of your queues’ settings in one table with filtering, search, and sorting. See queue settings
- Clone and adopt settings: Clone an existing queue along with its settings to create a new one, or adopt another queue’s settings into an existing queue.
- Queue favorites: Star frequently used queues and filter queue lists to favorites for faster navigation.
Review Experience
Reviewers have richer context and more control while working a job.- Job activity feed: See a chronological activity feed on each job, capturing assignments, decisions, and other events from both humans and agents.
- Internal notes and comments: Leave internal notes and comments on a job for other reviewers, directly in the activity feed.
- Edit communications before sending: When a workflow would send a communication on decision, reviewers can preview and edit the message before it goes out.
- Communication threads: Related communication messages are grouped into threads for easier follow-up.
- Long-running jobs: Jobs can wait for more information from a user and resume later, instead of being forced to a decision prematurely.
- Point Assignments in Review: Queue admins can show Point Assignments on the Review page for queues where point context helps reviewers.
- Saved job chat transcripts: Job chat transcripts are saved at decision time and viewable read-only on closed jobs.
- More readable details: Long IDs, URLs, activity titles, and wrapped policy badges now display more cleanly across review and job activity surfaces.
Quality Assurance
QA gives teams better ways to measure and reconcile review quality.- Forms in QA: Attach forms to QA queues, submit form answers alongside the grade, view submitted answers on the completed grade, and include form submissions in QA exports.
- More detailed QA scoring: QA decisions can show fractional accuracy and per-attribute breakdowns, making it easier to identify exactly where reviewers align or disagree. Learn more about QA
- Accuracy in QA exports: QA decision exports now include accuracy for downstream analysis. Export QA Decisions
- Archived policies in QA jobs: Archived policies now display on QA jobs so historical decisions stay readable.
- Reuse previous QA settings: Create a QA queue from a previous configuration instead of starting from scratch.
NCMEC Reporting
Generate and submit NCMEC reports from your workflows, with templates that adapt precisely to your data model.- Submit NCMEC reports from workflows: A dedicated workflow node can create and submit NCMEC report submissions as part of your automated child safety workflows. Learn more about NCMEC reporting
- Template-level entity overrides: Configure incident, user, media, phone, IP, additional information, and relationship mappings directly in NCMEC templates.
- Reportable policy overrides: Set reportable policies and match reported media on policy in template entity overrides so child safety reporting matches the way your policies are organized.
- Sharper report filtering: Filter NCMEC reports by decision entity ID.
Analytics and Exports
Analytics dashboards and exports support more precise operational reporting.- Faster, more customizable dashboards: Analytics has been rebuilt on a new data foundation for more reliable, customizable dashboards, now rolling out to customers. Build analytics dashboards
- Dashboard timezone control: Choose the timezone used for dashboard ranges, chart queries, and time-series labels so teams see consistent results across regions.
- Advanced chart filters: Use nested AND/OR filter groups in the chart builder for more precise dashboard cards.
- Current job state reporting: A new data source lets you build charts on the latest snapshot of your jobs, filtering and breaking down by current status, priority, and time-to-resolution.
- Richer job exports: Include full entity data when exporting jobs, bringing job exports closer to decision exports for downstream analysis. Learn more about data exports
- Decision and report exports: QA decision exports now include accuracy, and reports support queue and job-status filters with CSV export. Export QA Decisions
Connected Apps (MCP)
Connect approved AI clients to Cinder through MCP and manage access from Settings.- MCP setup in Cinder: Users with MCP access can connect Cursor, Claude, Codex, or a custom client from Settings > MCP. Learn more about MCP Connected Apps
- Admin controls: Admins can review registered apps, inspect organization grants, disable clients, and revoke user grants from the same settings area.
- More actions over MCP: Clone jobs into training queues and provision or invite users programmatically through MCP.
Webhooks and APIs
Engineering teams have more options for keeping downstream systems in sync.- Job closed webhooks: Subscribe to job closed events when a job is resolved.
- Decision review visibility: Decision review webhooks now appear in webhook tables and include accuracy in their payloads. Decision reviewed webhook
- Prediction scores in webhook payloads: Webhooks can include the numeric prediction score alongside the confidence band, helping downstream systems apply their own thresholds.
- Encrypted custom headers and egress IPs: Add encrypted custom headers to webhook endpoints, and view Cinder’s webhook egress IPs for allowlisting.
- Job API improvements: Create job comments through the API, filter jobs by multiple statuses in a single request, and pull the next job in a queue from a REST endpoint.
- User provisioning: Invite and provision users through the API, with SSO-aware invite emails when your organization enforces SSO.