The Endpoint-to-Cloud Privilege Security Checklist: 21 Controls to Eliminate Standing Access

PAM secured the endpoint; privilege moved on. Work through these 21 controls to find out where standing access is accumulating in your cloud, SaaS, and AI estate — and how to shut it down without slowing anyone. 

For twenty years, privileged access management (PAM) was an endpoint discipline: vault the credentials, strip local admin rights, broker and record the sessions. It worked because privilege lived where the controls did.

Today, infrastructure runs in AWS, Azure, and Google Cloud, crown-jewel data sits in SaaS, and a fast-growing fleet of service accounts, pipelines, and AI agents holds always-on permissions no vault has ever seen.

Attackers have noticed: they no longer break in, they log in — and inherit every standing entitlement that identity has accumulated. 

Use the checklist below to audit how far your privilege programme actually extends.

It runs in five phases, from the endpoint baseline you probably already have to the just-in-time (JIT) model that eliminates standing access entirely.

Treat every unchecked box as a finding: give it an owner, a deadline, and a metric. The phases are sequential by design discovery before right-sizing, right-sizing before JIT but nothing stops you piloting JIT on one high-risk system while the wider inventory work runs. 

WEBINAR SPOTLIGHT : 
Beyond the Endpoint: Where Privilege Went Next, and How Entitle Follows 
BeyondTrust's APJ Tech Talk Tuesday session walks this exact journey live, with Entitle discovering, right-sizing, and granting cloud access just-in-time. Reserve your seat → 

The 2026 Reality Check

The endpoint threat model hasn’t gone away — it has been joined by a much larger one. Recent industry research shows just how far standing privilege has spread beyond anything a traditional vault was built to see: 

• Machine identities now outnumber humans by roughly 80–100 to 1 in the average enterprise — service accounts, API keys, pipelines, and AI agents that mostly sit outside vault and directory visibility entirely.

• 92% of security leaders say they lack full visibility into their AI agent identities, and most doubt they could detect or contain a compromised one.

• Just 0.01% of non-human identities control an estimated 80% of cloud resources — a concentration of blast radius that rarely surfaces in a standard access review.

• 80% of security incidents still involve a compromised or misused privileged credential — the endpoint-era problem, replaying itself in the cloud.

What Standing Access Actually Costs

These aren’t abstract risks. IBM’s 2026 Cost of a Data Breach Report, based on 602 breached organisations worldwide, puts the global average breach at a record $4.99 million, up 12% in a year and the industries carrying the most standing entitlement pay the most for it: 

• Healthcare — $6.64 million per breach, the costliest sector for the 13th consecutive year, driven by standing access to patient PII.

• Financial services — $6.29 million, second-highest and a target for AI-driven attacks against core banking and trading access.

• Technology and industrial — $5.50 million each, where production and CI/CD access most often stands open by default.

• Energy — $5.2 million, with critical-infrastructure operators facing the heaviest concentration of AI-enabled attacks.

• United States — $11.5 million average, more than double the global figure.

Source: IBM Cost of a Data Breach Report 2026 (29 Jul 2026) 

Two of 2026’s highest-profile incidents show exactly how standing access turns into headlines: 

• Vercel / Context.ai, April 2026 — an employee’s one-time “Allow All” OAuth grant to a third-party AI tool became a standing credential attackers rode weeks later, after the vendor itself was compromised. The result: exposed API keys, source code, and roughly 580 employee records.

Source: Trend Micro, “The Vercel Breach: OAuth Supply Chain Attack” 

• Salesloft Drift, August 2025 — long-lived OAuth tokens tied to a single chatbot integration stood open across more than 700 organisations’ Salesforce environments. No malware, no password theft — just standing tokens nobody had time-boxed, used to run bulk exports until someone noticed.

Source: Cyber Security News, “Salesloft Takes Drift Offline After OAuth Token Theft” 

The Fix: Five Countermeasures, in Order

Neither breach needed a zero-day. Vercel and Salesloft/Drift both trace back to the same root cause: a grant that made sense on the day it was issued and was never revisited, reviewed, or time-boxed after that.

Standing access doesn’t fail loudly — it just waits for the day someone else finds it first. The five countermeasures below are what close that gap, and they only work in order: 

• Re-confirm the endpoint baseline — vaulted credentials, brokered sessions, no standing local admin. (Phase 1)

• Discover where privilege actually lives now — IaaS entitlements, SaaS admin roles, OAuth grants, non-human identities, AI agents. (Phase 2)

• Right-size and kill dormant grants — revoke what usage data says nobody needs, starting with birthright access to sensitive scopes. (Phase 3)

• Replace standing grants with just-in-time access — request, approve, provision, and auto-revoke, evidenced end to end. (Phase 4)

• Prove it continuously — measure usage fraction and leaver clean-up as ongoing metrics, not one-off audits. (Phase 5)

Work through the checklist below phase by phase — each one gives you the specific controls behind the summary above. 

Phase 1 — Lock Down the Endpoint Baseline

None of the cloud work matters if the original battleground is still open. Confirm the fundamentals first: 

• Remove standing local admin rights — deploy endpoint privilege management (EPM) so applications are elevated on demand and users never hold blanket admin, breaking the foothold-to-lateral-movement chain.

• Vault and auto-rotate every privileged credential — root, domain admin, and service-account passwords belong in a vault with automatic rotation — never in scripts, spreadsheets, or CI variables.

• Broker and record privileged sessions — every administrative session should pass through a gateway that enforces policy and produces a reviewable recording.

• Make least privilege the default posture — access is denied unless granted, and every elevation is time-bound and logged.

Phase 2 — Map Where Privilege Actually Lives Now

You cannot right-size what you cannot see. This phase is pure discovery — and it is where most programmes find their blind spot: 

Figure 1: Privilege has migrated from the vault-governed endpoint into IaaS entitlements, SaaS admin models, and non-human and AI identities — where it sits as unmanaged standing access. 

• Inventory IaaS entitlements continuously — each cloud platform exposes thousands of granular permissions composed through roles, policies, and inheritance; effective access must be computed, not guessed.

• Enumerate SaaS admin roles and OAuth grants — Microsoft 365, Salesforce, GitHub, and Snowflake each carry proprietary admin models that live entirely outside your directory and your vault.

• Catalogue every non-human identity — service accounts, API keys, CI/CD pipelines, and workload identities now vastly outnumber people in most estates — some benchmarks put the ratio above 80:1 — and most are rarely rotated or reviewed.

• Bring AI agents into scope — agents granted broad standing permissions so they “just work” hold privilege around the clock and act at machine speed when abused; most security teams admit their current IAM tooling was never built to govern them.

• Flag shadow-admin paths — a modest-looking permission — the right to modify a role or reset a credential — can quietly confer full control; cloud infrastructure entitlement management (CIEM) tooling exists to surface exactly these.

Phase 3 — Eliminate Standing Access

Discovery becomes risk reduction only when dormant privilege starts disappearing: 

• Right-size against real usage — compare granted permissions with what was actually used in the last 90 days, and revoke the difference — every dormant entitlement is attack surface with zero business value.

• Kill birthright grants to sensitive scopes — production databases, IAM administration, and tenant-wide SaaS roles should never be part of a default access package.

• Extend offboarding to cloud and SaaS — disabling the directory account is not enough; a leaver’s cloud entitlements and OAuth authorisations must die with them, in hours rather than weeks — exposed cloud credentials are routinely found and abused within minutes of leaking.

• Replace rubber-stamp reviews with usage-based certification — reviewers cannot interpret thousands of raw permissions, but they can act on “granted, unused for 90 days.”

Wondering how many of these boxes your organisation could tick today? The upcoming BeyondTrust Tech Talk includes a practical maturity assessment across the endpoint-to-cloud privilege spectrum — register here. 

Prefer to talk it through with someone first? Book 20 minutes with an Entitle specialist → 

Phase 4 — Operationalise Just-in-Time Access

The endpoint lesson, applied to the cloud: do not manage standing privilege better remove it, and grant access only when needed, for only as long as needed.

That model is zero standing privileges (ZSP), and it lives or dies on workflow. More than two-thirds of security teams already say they plan to deploy JIT access for privileged users — the gap is rarely intent, it’s the workflow to run it at scale: 

Figure 2: The just-in-time access lifecycle — every grant is requested in context, time-boxed at birth, revoked automatically, and evidenced end to end. 

• Take requests where people already work — engineers should ask for access from Slack, Teams, or the CLI — a JIT workflow that forces a portal detour will be bypassed.

• Automate policy-based approvals — pre-approved, low-risk combinations are granted in seconds; sensitive requests route to the resource owner for one-click approval.

• Time-box every grant and revoke automatically — the entitlement is provisioned natively in the target platform with an expiry attached — no ticket queues, no orphaned permissions.

• Evidence the full lifecycle — request, approval, grant, and revocation land in an audit trail that maps cleanly to compliance frameworks.

• Extend ZSP to non-human identities — workloads and AI agents should draw short-lived, scoped credentials on demand, exactly as humans do.

Phase 5 — Prove It

Three questions tell you whether the programme is real: 

• Can you enumerate every identity that can reach production data — human or machine, across cloud and SaaS, on demand — not after a two-week audit?

• Do you know your usage fraction — what share of granted access was actually exercised in the last 90 days — and is that number shrinking?

• Can you prove a leaver loses everything within hours — including cloud entitlements and OAuth grants that never touched the directory?

Score yourself honestly. Most organisations clear Phase 1 and stall in Phase 2 — not for lack of tooling, but because visibility and workflow were never extended beyond the endpoint.

The encouraging news: the second half of the journey moves faster than the first, because once entitlements are discovered and usage is measured, converting standing grants into just-in-time access is largely an automation exercise.

Privilege has left the endpoint for good; the discipline that secured it does not have to stay behind. 

Entitle Use Cases by Industry

Where teams are putting this into practice today — by sector, and cross-industry: 

FINANCIAL SERVICES : Segregation-of-duties enforcement — time-boxed access to core banking and trading systems so no single standing grant crosses an audit boundary. 

HEALTHCARE : Need-to-treat EHR and PHI access — clinicians and support staff get record access for the duration of care, with the time-stamped audit trail HIPAA reviewers expect. 

TECHNOLOGY & SAAS : JIT production access — engineers request from Slack, Teams, or the CLI; access is provisioned natively with an expiry attached. 

RETAIL & E-COMMERCE: Narrow, expiring access to PCI-scoped systems — seasonal staff and contractors get payment-system access that disappears the moment the shift or contract ends. 

ENERGY & CRITICAL INFRASTRUCTURE : Time-boxed vendor and OT remote access — third-party maintenance access is revoked automatically the moment the work order closes. 

CROSS-INDUSTRY 

• Temporary admin escalation & break-glass access — grant emergency privileged access for a fixed window, then auto-revoke.

• On-call access — permissions tied to the rotation, not the person, so access disappears automatically when the shift ends.

• Usage-based access reviews — replace rubber-stamp certification with “granted, unused for 90 days” signals reviewers can actually act on.

• Non-human and AI agent access — short-lived, scoped credentials for service accounts, pipelines, and agents, governed on the same lifecycle as human grants.

See all Entitle use cases →

SEE IT LIVE, OR TALK IT THROUGH 

Beyond the Endpoint: Where Privilege Went Next, and How Entitle Follows 

A practical APJ Tech Talk Tuesday session from BeyondTrust showing every phase of this checklist in action discovering, right-sizing, and granting cloud access just-in-time with Entitle on the Pathfinder platform. Built for security and IAM teams with established PAM programmes, organisations midway through their identity security journey, and cloud and DevOps teams eliminating standing access without slowing delivery. 
1. Register free — Reserve your seat → 
2. Talk to an Entitle expert — skip the slides and walk through your own environment 1:1. Request a demo →