New permission sets, SSO controls, and a unified agent audit log let admins govern human and AI activity in Airtable from one place.
This week's launch system was built and run by a mix of humans and AI agents — Claude and ChatGPT creating fully functional Airtable bases, complete with tables and interfaces, team members filing work throughout with natural language prompts, and Canvas turning live data into a presentation. Today's release is what makes it all trustworthy at scale: data permissions for integrated agents, SSO and sharing controls, and a single agent audit log that records every action, human or agent, in one queryable place.
Key takeaways
- The agent audit log attributes every action to its actor — human or agent — in the same record, filterable by actor, base, event type, org unit, and date.
- New permission sets (in beta), Airtable SSO controls, and sharing controls govern who and what can touch a given base, including connected AI agents.
- Admin persona packs give admins ready-made prompts for recurring governance tasks: tracing a record's full history, running a weekly access review, checking scope for a user or agent, and sweeping permissions clean after an offboarding
What's new in Airtable admin and governance
Airtable is introducing permission sets (in beta), SSO and sharing controls, AI for HIPAA, and an audit log that extends Airtable's existing enterprise audit logging to cover agent activity specifically. Together, these give admins the same level of control over AI agents that they've always had over human collaborators.
Permission sets, SSO, and sharing controls: Governing who touches the launch base
Permission sets (in beta) extend Airtable's existing access model to explicitly account for connected AI agents. The same "who can reach this base, and under what conditions" question admins have always answered for people, now answered for agents too. This builds directly on Airtable permissions you've already configured; nothing about how Airtable permissions work for your human collaborators changes.
Keeping Claude connections inside your org
Once a team has spent a week connecting Claude to company data — as builders did on Monday and end users did on Tuesday — this is how admins make sure that connection only flows to the organization's Claude Enterprise account, not a personal one. This is the Anthropic verified-domain connector restriction.
The agent audit log: Every action, on the record
A whole week's activity in one queryable record — Monday's agent-created tables, Tuesday's form submissions and interface reads — every action attributed to its actor, human or agent, side by side (e.g., "Claude — Riley Yu" next to "Riley Yu"). Filterable by actor, base, event type, org unit, and date; exportable via the audit API to a SIEM.
The Airtable audit log itself isn't new. Its 180-day retention, admin-panel and API access, and CSV/JSON export capability is already documented. What's new is agent attribution living inside that same log — Claude and ChatGPT showing up as actors next to the people who authorized them.
Admin persona packs: Trace a record, review the week, check scope, sweep an offboarding
Ready-made prompts and workflows cover the governance tasks that come up every week: tracing a single record's full history, running a weekly access review, checking scope for a given user or agent, and sweeping permissions clean after someone leaves. An admin doesn't have to write these from scratch under pressure during an incident, a compliance request, or someone's last day, the prompt for that exact situation is already there.
AI for HIPAA: bringing AI into regulated workflows (early access)
HIPAA workspaces keep regulated data like health records inside strict compliance boundaries. AI for HIPAA extends that same boundary to AI: only AI capabilities that meet HIPAA requirements are available inside the workspace, and only from model providers who've signed the required data-protection agreements. That means the same MCP-driven building, Canvas dashboards, and agent workflows covered earlier this week can now run inside a regulated workspace too, without a team having to choose between using AI and staying compliant. AI for HIPAA is early access. Interested teams should talk to their Airtable account manager.
Move fast without losing the paper trail
Governance here isn't what slows the rest of this week's launch system down. It's what makes it safe to let it move fast in the first place. The same agents that built a launch tracker on Monday and took in requests on Tuesday are now fully on the record, right alongside the people who directed them.
This is Day 4 of a five-day series. Catch up on Day 1: Interfaces and automations with Airtable's MCP server, Day 2: Airtable MCP for teams, and Day 3 on Airtable Canvas.
Don't miss day 5
Frequently asked questions
Airtable's audit logs already track actions with detail like actor, timestamp, and object affected. What's new is that agent-driven actions are now attributed alongside the human who directed them, in the same log, rather than appearing as anonymous API activity.
Yes. Audit log events can be retrieved via Airtable's API and connected to a third-party SIEM tool. See the Audit Log Integration Guide.
No. Per Anthropic's own documentation, this setting is a safeguard against accidentally connecting a work account to the wrong Claude account — it isn't a data loss prevention control against deliberate data movement.
No. It's early access, with availability starting mid-August. Interested teams should talk to their Airtable account manager.
Airtable’s audit log events are retained for 180 days.
