Has anyone had any luck developing an AI chat, placing it in the JSM portal, and making it compliant with Article 50 of the EU AI Act?
We use JSM to serve external customers (CSM still feels underdeveloped).
Good day Jaroslaw To comply with the AI Act, the simplest approach is to use the native Atlassian Intelligence Virtual Agent (since it already covers all necessary adaptations). If you build your own, the initial greeting must clearly inform the user that they are talking to an AI. You can review the privacy and compliance details in Atlassian's transparency notes and the AI and data policy. Hope this helps ; )
For a native-first design, start with the JSM virtual service agent in the help center, make the bot clearly identifiable as AI in the greeting, provide an obvious human-agent escalation path, and keep a record of the interaction and handoff. Limit its knowledge sources to approved, current articles and add a fallback when confidence or article coverage is low. Then have your privacy/legal team map the configuration, notices, retention, and customer disclosures to the requirements that apply to your organization—Article 50 compliance is not something the portal setting alone can certify. Test the anonymous and authenticated portal journeys separately before enabling it for external customers.
@Josh @Jairo Merea Thank you both for your responses. One of the requirements of the act is to introduce a launcher/banner that prominently warns the user that they are about to interact with an AI chat. From what I understand, these are not achievable by default in the portal, as users can be informed about interacting with the AI chat once they activate the window and are greeted by the agent. Another cornerstone is chat accessibility, i.e. Ensure screen-reader compatibility, colour-contrast compliance, and localisation where relevant. Which I also understand is not available in the out-of-the-box solution, as Atlassian does not test its individual elements for WCAG/EAA but tests its products as a whole, if that makes sense.
When we deployed Rovo chat, none of the above produced more than partial compliance, hence my question.
For the disclosure part, one workaround that doesn't depend on the chat itself: JSM lets you set an announcement on the help center and on each portal (portal settings > announcement). A short line such as "This portal includes an AI assistant; you can ask for a person at any time" is visible before anyone opens the chat window, which seems closer to what the Act asks for than the greeting inside the chat.
For accessibility, the reference point is Atlassian's own conformance report (ACR/VPAT) for Jira Service Management, linked from atlassian.com/accessibility. If the chat widget isn't covered there, I'd test it directly with a keyboard only, a screen reader (NVDA or VoiceOver) and a contrast checker, write down what fails, and raise it as a JSM issue so it's on record. That also gives you something concrete to show if anyone asks how you assessed it.
Disclosure: I work on accessibility tooling for Confluence (Clearwren), not on JSM, so treat the second part as a testing approach rather than a known answer.
It looks like you're new here. Sign in or register to get started.
