Privileged Access, a runtime broker now in public preview, and a planned Snowflake integration each attack standing access from a different direction: infrastructure permissions, task-time issuance, and third-party agent trust. Only the first two are available now. The test is whether they remove standing access or add another control plane.
Standing access is the finding every identity audit turns up and no one remembers granting. An engineer gets a role for a migration. The migration ends. The role stays. A service account picks up a permission to unblock a deploy, and nobody takes it back. 1Password built Privileged Access, moved a runtime broker into public preview, and announced a pending Snowflake integration around one mechanism: grant access for the task, then revoke it when the task ends.
The company's AI Agent Governance survey puts a number on that failure. Forty percent of developers grant AI agents persistent access to systems, meaning the access is never revoked after the job is done. That figure is the baseline a CIO can measure against once Privileged Access is running.
Privileged Access extends 1Password's Unified Access platform into privileged access management. It provisions permissions in the target system's own policy layer instead of replacing the tools a security team already runs. The product is built on Apono, the just-in-time access company 1Password acquired in June. CEO David Faugno put the design rule in one sentence: access must be granted for a specific task, then revoked when that task ends (1Password, 2026).
The mechanics matter more than the positioning. Discovery scans cloud, database, and Kubernetes environments for access that accumulated and was never reviewed. Provisioning happens at request time and is scoped to the task. Every request, approval, and access event is logged with attribution that can support SOC 2, HIPAA, PCI-DSS, ISO 27001, and GDPR evidence. Just-in-time access and zero standing privilege predate this product. 1Password applies them from the start to non-human identities, with agents named beside employees and service accounts in its own framing.
The runtime broker targets a narrower gap: long-lived values copied into pipelines instead of requested at run time. It left private beta and entered public preview for GitHub Actions, issuing access for a single workflow run rather than leaving static values in configuration files. GitHub's Ben De St. Paer-Gotch, Director of Product Management, described the goal as removing those values from pipeline configurations to cut theft risk (1Password, 2026).
General availability has no announced date.
Teams evaluating the product today can test only what is shipping: GitHub Actions and custom OpenID Connect configurations. Machine workloads and AI agents remain on the roadmap. Scope a pilot to GitHub Actions delivery, the only workload shipping today.
Most access systems treat agents like human employees or service accounts, and agents behave like neither. The access issued to them tends to outlive the task.
Jason Meller, VP and Security Strategist at 1Password, states the structural problem: security has spent two decades asking people to slow down, and people have spent two decades saying no. If the secure path is not the fast path, the secure path will not be used.
The Snowflake work applies the same model to a different problem: third-party agents reaching into Snowflake from outside it. 1Password named the integration with its July launch and scheduled general availability for the second half of 2026. Snowflake's Black Hat 2026 announcement puts the work inside Cortex AI Gateway. Most of those agent-governance controls, including third-party identity integrations, are listed as private preview (Snowflake, 2026). Mayank Upadhyay, Snowflake's Chief Security and Trust Officer, described the aim as a secure foundation for agent interoperability (1Password, 2026). That direction matters as more vendors' agents call into shared data platforms with no common trust layer.
Enterprises that run both platforms should track the roadmap now. A partnership scheduled for later is a different evaluation question than a product already provisioning access in production.
The survey sample is 500 IT and security professionals and 500 developers at U.S. firms with at least 250 employees. That is large enough for the headline figures. It does not show whether the 40 percent persistent-access finding holds at smaller companies or outside the United States. 1Password has not published a breakdown by company size. It has also not disclosed when the public preview becomes generally available, or how fast the Snowflake integration will move beyond its first scope. All survey figures are vendor-supplied and unaudited.
Ask 1Password for the before-and-after standing-access count from a live Privileged Access deployment. If the vendor cannot produce one, the survey baseline is all it has.
Sources
1Password. "1Password Launches Privileged Access to Eliminate Standing Access Across Human and AI Identities." 1Password, 28 July 2026, 1password.com.
1Password. "1Password Joins Snowflake to Tackle AI's Next Security Frontier: Trusted Agent Interoperability." 1Password, 28 July 2026, 1password.com.
Atwell, Elaine. "1Password's Research Finds AI Agent Adoption Is Outpacing Governance." 1Password, 28 July 2026, 1password.com.
Janardanan, Radhika, and Nitika Gupta. "Snowflake Launches Cortex AI Gateway and Advanced AI Security at Black Hat 2026." Snowflake, 28 July 2026, snowflake.com.
