Permissions and control

A skill can never do more than you could do in Sessionboard yourself. That is not a promise the skill makes: it is enforced by Sessionboard on every call, whichever assistant sends it.

Every action runs as you

You sign in, not a key. The assistant connects with OAuth: you sign in to Sessionboard, pick the organization and choose Authorize. There is no shared API token.
Your role applies. Each call is checked against your Sessionboard role and permissions, the same as the screens you use: what you can see in Contacts, the Awards programs you manage, the Learning permissions you hold, whether you can change an event. A skill that reaches something your role cannot gets the same refusal you would.
Your scopes apply. At sign-in you choose what the assistant may do. Without write access it can read but not change anything, whatever the skill says.
One organization per connection. The assistant works in the organization you authorized and nowhere else. Regions are separate: a connection to the US server cannot see a European organization.

It asks before it changes anything

Every skill reads first. Before it creates, changes, publishes, sends or deletes anything, it shows you exactly what will change (names, counts, recipients, dates) and waits for your yes in the conversation. It never takes an earlier answer or a plan as that yes.

Assistants add their own check: tools that change data are marked as such, so Claude and ChatGPT ask you before they run one.

Some actions wait for a person in Sessionboard

For the actions below, saying yes in the chat is not enough. Sessionboard answers that it needs approval and gives a link; a person with the right permission opens it in Sessionboard and approves it, and it then runs as them:

opening a call for speakers and publishing an agenda
opening Awards nominations, inviting Reviewers, reminders and announcing Winners
publishing courses to learners, changing what learners pay, and sending credit to CE Broker
launching campaigns, sending presenter kits and engagement emails, publishing pages, libraries and Sites, approving ad spend, and making share links for results and sponsors

Choosing Awards Winners is never done by an assistant: a person does it in Sessionboard.

Other changes run once you say yes in the chat, with no second approval: for example editing an agenda that is already published (the website shows the change at once) or sending a contract for signature. Each skill’s page lists exactly which of its steps wait in Sessionboard, under Where it stops for you.

What administrators control

AI features for the organization. When AI is off, no assistant can connect. An organization admin can check under Organization Settings → API Tokens: a banner reads AI features are disabled for this organization when it is off. Contact your account manager to change it.
Who can connect. Only roles that include Connect AI assistants can authorize an assistant. Organization administrators have it; grant it to other roles in your organization’s role settings.
What each person can do. Roles and permissions in Sessionboard decide what a connected assistant can read and change, so tightening a role tightens the assistant too.
Attendees. The attendee connection, which the attendee skill uses, is off for every event until you switch it on, and it can only read the attendee’s own event.
Claude for your whole team. On Claude Team and Enterprise plans, an Owner decides whether skills are allowed and can add Sessionboard skills for everyone (see Install skills).

What is recorded

Changes an assistant makes through the Marketing, Awards, Learning and CRM tools, calls for speakers and agendas, and event pages and emails are recorded in Sessionboard with the assistant’s name and the person who asked, beside changes made on screen.