Zapier automation security guide covers OAuth and admin controls
Zapier's updated guide says teams can build automated systems from AI tools or the platform itself, with OAuth-managed connections, admin controls and logged runs.

Key takeaways
- Zapier's guide, updated in November 2026, says users can build full systems from an AI tool or inside Zapier, across more than 10,000 apps.
- Zapier says requests route through its platform, so AI tools are not exposed to credentials, and connections are OAuth-scoped and revocable.
- Zapier Tables stores workflow data, and AI fields can draft content that a person reviews before a Zap sends it.
- Workflows can run automatically or only when clicked, with each run logged, according to Zapier.
In this article
What the guide covers

Zapier has updated its guidance on Zapier automation security, describing how teams can link apps, stored data and AI agents into one system while keeping credentials away from those agents. The post was first published in August 2024, most recently updated in November 2026, and dated 18 November 2026 on the company blog. It is vendor material, so its safety claims come from Zapier itself and have not been independently checked.
According to Zapier's blog, the aim is to move beyond simple triggers, where one event causes one action, towards complete systems. The post lays out three steps: connect Zapier to what it calls an agent harness, meaning the AI tool a team already uses; store and review data in Zapier Tables; then act on that data with Zaps, the name for its automated workflows. A user describes what should happen, an agent drafts the Zap and deploys it, and Zapier says the workflow starts when something happens in a connected app, runs the same way each time, and uses AI only for steps that need judgement. One-off tasks are also possible, such as adding a task to a project tracker while working in Claude, or sending a Slack message from ChatGPT.
How credentials are handled
Zapier says every request passes through its platform, so the AI tool is not exposed to the secret keys behind those connections. Connections use OAuth, a sign-in standard that grants limited access without sharing passwords, and are scoped to particular apps and actions. Scoping means each connection allows only named actions rather than open access to an account. Connections can be revoked from a single place, and an IT administrator can decide what each AI tool is allowed to touch. These are Zapier's own statements, and the material contains no independent test results.
Where the data lives
The guide points to Zapier Tables, a database built for automation, as the place to keep information produced by automated processes. A Zap can send data to a table from any connected app, such as new leads from a customer relationship management system or requests from a project tracker, so a team can store, track and review everything in one place.
Tables also include AI fields that can draft content, for example emails to leads. A person reviews the draft, then a Zap sends it with one click. Zapier says AI by Zapier can add steps that analyse, categorise or decide the next move on their own, and that the platform connects to more than 10,000 apps.
Automatic runs and manual checkpoints
By default these processes run on their own: when a new lead arrives, the Zap fires at once. Zapier says a workflow can also become a button inside a table, running only when someone clicks it, which gives teams a manual checkpoint. Each run is logged, and if something breaks Zapier suggests a fix, according to the post. Run logs matter because they leave a record to check when a workflow fails. Examples given include notifying a sales team in Slack, adding prospects to a CRM, sending onboarding assignments to new staff and confirming approved software requests.
How to read the claims
The guide is product guidance as much as news, so its security claims should be read as Zapier's description of its own design rather than as verified findings. Organisations considering such a setup should consult Zapier's official documentation for details on OAuth scopes, admin controls and run logs before relying on them. The practical takeaway is that automation can run fully on its own, wait for a manual click, or mix both, with logging and revocation presented as the main safeguards.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



