Halo has shipped its second release of 2026 (version 2.236), and it's a big one — improvements across AI, approvals, access control, reporting and the self-service portal. Most "what's new" write-ups just list the features. We work with Halo every day, so here's the version that matters: what's genuinely useful, and what it changes for how you run a service desk.
AI stops being a chatbot and starts doing the work
The headline is AI, but not in the way most vendors mean it. Two new building blocks — AI Abilities and AI Agents — can now be dropped straight into integration runbooks as a step type.
An AI Ability is a set of instructions you send to the AI mid-automation; its response feeds back into your runbook variables, and can be forced to clean JSON so a single call can populate several fields at once. An AI Agent goes further: it authorises as a Halo application with its own identity and permissions, and can carry out a sequence of tasks — retrieving ticket data, searching the knowledge base, adding actions to tickets — using the same functions exposed through Halo's MCP server.
In plain terms: AI is no longer a bolt-on assistant. It's a first-class step inside your automations, running under controlled permissions. That's the difference between AI that suggests and AI that acts.
Alongside it, ticket clustering now groups similar tickets automatically based on their embeddings — each cluster gets an AI-generated description and resolution, and you can track cluster volumes over time. It's a quiet feature with real weight: it surfaces emerging major incidents and your most common ticket types without anyone building a report.
Approvals grow up: parallel steps and risk-based routing
Approval processes can now run steps in parallel by grouping them. The group approves once every step clears its threshold, and rejects the moment any one step hits its rejection threshold. And Risk Score is now available as a criterion for approval routing, so a change with a high calculated risk can be sent down a stricter approval path automatically.
For anyone running proper change management, this is the kind of behaviour you previously had to bend the tool to achieve. Now it's native.
Tighter control, for the people who care about it
Role-based access control has been extended to Active Directory configuration, approval groups, and event management rules — you can now lock down who can view and change the automations that run your environment. There's also a new Security Questions module for call handling: agents can verify a caller's identity in real time before anything is logged against them. For regulated and public-sector teams, both land squarely on the audit and security requirements that tend to surface in procurement.
The self-service portal gets more usable
Several changes make the portal less of a chore for end users. Child tickets now nest under their parent in the "My Tickets" view, the service catalogue can surface a user's most recent and trending services, and users can raise a ticket directly from an asset record without re-entering the details. Small individually — but this is where end-user adoption is won or lost.
Also worth knowing
OLA breach notifications now work like the existing SLA thresholds. Runbook steps have an in-platform request and response log for faster diagnostics. Scheduled reports can be sent to distribution lists instead of hand-maintained email lists. And there are two new integrations — SailPoint IdentityNow and AssetBots.
What this means if you're weighing Halo up
Every module, and now this full sweep of AI capability, is included in Halo's single all-inclusive licence — no per-feature upsell, no separate "AI tier". If you're comparing Halo against a platform where AI agents, parallel approvals and risk-based routing arrive as paid add-ons or premium SKUs, 2026.2 widens that gap.
We implement this for a living. If you want to know what these features look like configured for your environment — or you're mapping a move off ServiceNow and want to see how the AI and approvals stack up — get in touch.