For years, authorisation was the quiet corner of security: a few role names, a spreadsheet, and a hope that nobody asked hard questions. The picture changed noticeably over the past eighteen months. Standards matured, open-source engines reached new stages of community governance, vendors consolidated products, and research kept returning the same uncomfortable numbers about neglected credentials. This review pulls together the evidence on where IAMOps stands today, treating it as the informal label the Ops Atlas gives to authorisation as code.
The risk signal has not softened
Start with the problem statement. OWASP’s 2025 ranking again put broken access control at the top, noting that all of the applications it examined showed at least some weakness of that kind. It is a striking result, though it describes a sample rather than the internet at large.
A monitoring vendor’s cloud-security report from October 2025 adds texture on credentials. Keys that had stayed active beyond twelve months belonged to 59% of IAM users on AWS, 55% of service accounts on Google Cloud and 40% of apps registered in Entra ID. Its authors also called roughly one outside integration in eight (12.2%) dangerously over-privileged. Different platforms, similar pattern: access gets created quickly and retired slowly.
Incidents still point at permissions, not only at entry
High-profile cases keep showing the same shape. In the 2019 incident that led to a 2022 conviction, a misconfigured web application firewall exposed role credentials, and the permissions on that role set the scale of what could be taken. In January 2024, a major software company disclosed that attackers had used an old test OAuth application with elevated access to grant themselves mailbox rights. In both stories the initial foothold was smaller than the permissions behind it. That is why practitioners now frame the question as blast radius per identity rather than only perimeter.
Engines and projects: consolidation and governance
The open-source policy landscape shifted in ways worth tracking.
- Open Policy Agent reached graduated status in the Cloud Native Computing Foundation in 2021 and remains the general-purpose reference for policy as code.
- Cedar entered the CNCF sandbox in October 2025, two years after Amazon released it as open source. A 2024 paper described machine-checked proofs of core properties, a rare level of assurance for a policy language.
- OpenFGA, a relationship-based engine inspired by Google’s 2019 Zanzibar paper, became a CNCF incubating project on 28 October 2025.
- SPIFFE and SPIRE, for workload identity, graduated in 2022, making short-lived workload credentials a mainstream option.
The trend is toward neutral governance and specialised engines: general rules in one family, fine-grained relationships in another, workload identity underneath. Teams no longer have to treat the choice as a vendor bet.
Standardising the question
Perhaps the most significant development for long-term interoperability is the OpenID Foundation’s Authorization API 1.0, approved as a Final Specification on 11 January 2026. It defines a common way for an application to ask a policy decision service whether a request is allowed. If it gains adoption, engines become swappable, much as databases became accessible through shared drivers. It is too early to say how widely it will be used, and readers should treat adoption claims with caution.
Products come and go
Commercial tooling is also in flux. Microsoft retired the standalone Entra Permissions Management service on 1 November 2025 and pointed customers to cloud infrastructure entitlement features inside Defender for Cloud. For buyers, the lesson is practical: keep your policy files and measurement habits portable, because the product wrapping them may change.
Analysis is increasingly metered, too. The Atlas snapshots AWS prices for October 2026: twenty cents a month for each role or user scanned for unused access, a fifth of a cent per custom policy check, nine dollars monthly per resource for internal-access analysis, and nothing for external-access findings or validation. A fleet of 850 roles therefore costs about $170 monthly, while a busy repository running five checks on each of 400 pull requests spends roughly $4. Continuous measurement is affordable for most teams.
What the data suggest about practice
Evidence from the Atlas’s maturity ladder, which is its own framework rather than an industry standard, gives a way to place organisations. It climbs from ad hoc, to guarded (no fresh long-lived keys), to codified (nine in ten policy edits flowing through CI), to measured (reviews every month), and finally to adaptive, where removals trigger no incidents. No public survey places organisations on this ladder, so treat it as a self-assessment tool rather than a league table.
Reading the trend lines responsibly
A few cautions apply. Vendor-published statistics measure their own customer base. Sampled percentages describe the sample. And the informal label means no tracker counts “IAMOps teams” as such. With those caveats, three directions appear reasonably well supported: credentials are shifting from stored keys to federation; policy is moving into repositories with automated checks; and measurement of unused permissions is becoming routine rather than exceptional.
The full tour of tools, standards and pitfalls sits in the Ops Atlas IAMOps entry, which cites the dates and versions used here.
Where this meets the operations conversation
Automation is a double-edged sword for permissions: software that acts on its own needs boundaries as much as people do. Readers weighing that tension will find useful background in the practical roadmap for operations teams adopting AIOps and in our look at resolving incidents faster with automated assistance, both of which involve systems acting with delegated authority.
Outlook
Expect the next year to bring more tooling that reads the shared decision API, more reports quantifying unused access, and a gradual rise in teams publishing their own utilisation numbers. Direction matters more than speed here: permissions are becoming measurable, versioned and testable, and the teams that adapt first will spend less time explaining surprising access to auditors.

