Cloud Security

6-Step Review: Find Exposed Buckets Before Attackers Do

November 5, 2026 · 5 min read · By CloudAI Security
6-Step Review: Find Exposed Buckets Before Attackers Do

To find exposed buckets before an attacker does, do not audit bucket permissions one at a time. The review that scales checks whether each cloud’s public-access guardrail is actually enforced at the account or organization level, lists only the storage that sits outside that guardrail, and confirms each remaining candidate with an unauthenticated request. That sequence turns an unbounded inventory problem into a short exception list one engineer can finish in a single afternoon, and it works the same way across AWS, Google Cloud, and Azure.

The whole method answers one question: are the settings that make object storage publicly reachable switched off, at the highest level each cloud allows? In AWS, that switch is S3 Block Public Access, whose settings override bucket policies and ACLs and apply as the most restrictive combination across organization, account, and bucket. Enforcement state, not per-bucket intent, is what an attacker actually encounters. This site already covers the monthly storage exposure drift signals that precede a leak and the runbook for containing a leaked Terraform state file; this review is the step between them, when exposure is still only a misconfiguration.

Start From Enforcement State

Every major provider ships a control that makes public access structurally impossible rather than merely discouraged. Reviewing that control first is what separates a periodic review from a full audit: you verify a handful of organization-level switches instead of sampling thousands of bucket policies. The table below is the entire scope of step one.

CloudGuardrail to verifyWhere it can be enforcedWhat counts as an exception
AWSS3 Block Public Access, all four settingsOrganization, account, bucket, access pointAny bucket or access point not covered by account-level or organization-level block settings
Google CloudPublic access preventionBucket setting, or organization policy constraint storage.publicAccessPrevention at project, folder, or organization levelBuckets reported as inherited while no constraint is set anywhere in the hierarchy
AzureallowBlobPublicAccess on the storage accountStorage account; enforceable with Azure PolicyAny storage account that still permits blob public access at the account level

The hierarchy is the reason this method works. A guardrail set at the top of the resource hierarchy keeps protecting resources whose local settings are weaker, so the only buckets worth reading policies for are the ones the hierarchy does not cover. That is your exception list, and it is usually small.

Triage Only the Exceptions

With the exception list in hand, pull the effective permissions for those resources only. On AWS, that means the bucket policy, ACLs, and access point policies for each unprotected bucket. On Google Cloud, Cloud Asset Inventory can return the IAM policies for every bucket in a project, folder, or organization in one query, which is faster than reading buckets individually. On Azure Resource Manager storage accounts, anonymous access to blob data is always prohibited by default, so check both the account-level anonymous access setting and any container configured for public read. If an exception exists for a legitimate reason, such as hosting public assets, record it in an exception register with an owner and a review date; an undocumented exception is indistinguishable from a mistake.

Verify With Anonymous Probes

Enforcement state tells you what the platform should block; a probe tells you what it actually blocks. From a context with no credentials, issue a plain HEAD or GET request against representative object URLs in each storage resource on the exception list and read the response. When public access prevention is enforced, requests authorized with allUsers or allAuthenticatedUsers fail with an HTTP 401 or 403 status code. The probe should return a refusal status for every URL, never object bytes. A probe from an unauthenticated context is the closest you can get to the attacker’s view without paying an external scanning vendor, and it catches caching layers, CDN overrides, and signed-URL leakage that policy reviews miss.

Track Drift Between Reviews

A one-off review decays immediately. New accounts, new regions, re-enabled public access for a migration, and Terraform applies that remove a guardrail all reopen the same hole. Keep the previous run’s guardrail-state export and diff it against the current one every cycle; any resource that moved from protected to unprotected is a higher-priority finding than a stable, documented exception. Tie the review to a fixed cadence with a named owner, and make the diff output itself the artifact your team signs off, so the review produces evidence rather than memory.

Six-Step Review Procedure

  1. Export guardrail state for every account, project, and subscription: S3 Block Public Access settings, public access prevention state, and allowBlobPublicAccess values.
  2. Diff the export against your approved baseline and list every storage resource that sits outside a guardrail.
  3. For that exception list only, pull effective policies, ACLs, and access point policies.
  4. Run provider detection tooling as a cross-check: IAM Access Analyzer for S3 flags buckets with bucket ACLs, bucket policies, or access point policies that grant public access, while Azure Policy compliance reports list storage accounts that still allow anonymous access.
  5. Probe each remaining candidate anonymously and treat any response that returns object bytes as an incident, not a finding.
  6. Close the loop: enforce the guardrail where possible, register the deliberate exceptions with owners and expiry dates, and archive the diff as the review record.

Sources