Security Posture

Your resources checked on every change against ISO 27001 and BSI C5: findings, accepted risks, signed reports.

Security Posture checks your resources on Deutsche Cloud against security controls all the time and tells you what fails, why it matters, how to fix it and who changed it. It maps every control to ISO/IEC 27001:2022 Annex A and BSI C5:2020, so the same checks are evidence for an audit.

It needs no setup and no tokens: the platform already knows every security group rule, bucket policy, database setting, patch report, access key, member and branch rule of your organization, and checks them from the inside. Console: Security & identity → Security Posture (/posture).

Evaluating your own resources is included; nothing in Security Posture is billed.

When it checks

  • On every change. Every successful write to the API is in your audit log. Security Posture follows it and re-runs the controls of the service that changed within seconds, so opening SSH to the world shows up as a finding before you have closed the tab.
  • After a scan. When the registry finishes scanning an image, or the Git service finishes scanning a repository's dependencies, the controls that read those results run again.
  • Every hour, everything: certificates expire, keys get old and new vulnerabilities are published without anyone changing anything.
  • On request: Evaluate now on the overview, dcloud posture evaluations start, or POST /posture/api/evaluations (at most once a minute).

What it checks

Control Service Severity ISO 27001 BSI C5
Security groups open only web ports (80/443) to the internet Compute High on a group attached to an instance with a public address, otherwise medium 8.20, 8.27 COS-01, OPS-23
Instances install security updates Compute High with security updates waiting for more than 7 days; medium when patching is off or the instance does not report 8.8, 8.19 OPS-18, OPS-22
Block volumes are encrypted at rest Compute Medium 8.24 CRY-03
Buckets are not readable by anyone on the internet Object Storage High 8.3, 8.12 PSS-09, IDM-07
Databases are not reachable from the internet Databases High 8.20, 8.22 COS-01, COS-06
Databases refuse connections without TLS Databases High when also public, otherwise medium 8.24, 8.21 CRY-02
Database storage and backups are encrypted at rest Databases Medium 8.24 CRY-03
Databases keep at least 7 days of backups and the last one is recent (26 h) Databases High (last backup), medium (retention) 8.13 OPS-06, OPS-07
Databases are protected against deletion Databases Low, medium for clusters 8.9, 8.13 OPS-06
Every database's backups were restored at least once in 180 days Databases Medium (younger databases are not applicable) 8.13 OPS-08
Every member signs in with a second factor Identity Critical for owners and administrators, high otherwise 8.5, 5.17 IDM-09
Access keys unused for 90 days are deactivated Identity High for organization keys, medium otherwise 5.18, 8.2 IDM-03, IDM-05
Access keys expire and are rotated within a year Identity Medium 5.17 IDM-08
Few principals can do everything (at most 4 with every action on every resource) Identity Medium 8.2, 5.15 IDM-06
The audit log is exported and kept for at least a year Identity Medium (retention), low (export) 8.15 OPS-12, OPS-14
Single sign-on, where set up, is enforced Identity Medium (not applicable without an enabled SSO connection) 5.17, 8.5 IDM-09, IDM-03
Images in use have no critical vulnerabilities that have a fix Registry Critical 8.8 OPS-18, PSS-11
Images in use were rebuilt within 90 days Registry Medium 8.8, 8.19 PSS-11
Images in use do not run as root Registry Medium 8.27, 8.9 OPS-23, PSS-11
Images in use are built on an operating system that still gets security updates Registry High 8.8, 8.19 OPS-18, PSS-11
The default branch takes changes only through a reviewed pull request Git High 8.4, 8.25, 8.32 DEV-03, DEV-09
Repositories have no open critical or high dependency alerts Git Critical or high 8.8, 8.28 OPS-18, DEV-01
No secret was pushed past secret protection and left unresolved Git High 8.12, 5.17 IDM-08
Pushes that add a credential are refused (secret push protection on) Git Medium 8.12, 8.25 IDM-08, DEV-03
Certificates in use are valid for at least 14 more days Load balancers, CDN, apps High (also when renewal failed), critical under 3 days 8.24, 8.21 CRY-02
Plain HTTP redirects to HTTPS Load balancers, CDN Medium 8.24, 8.21 CRY-02
Hosted zones are signed with DNSSEC DNS Low (only zones delegated to the platform) 8.21, 8.20 COS-01
Kubernetes API endpoints are not open to the whole internet Kubernetes High 8.20, 8.22 COS-01
Every control could be evaluated Security Posture Medium – –

"In use" for an image means tagged, not expired, and pushed or pulled in the last 30 days. Images scanned before the scan recorded the image's user and OS support status show "not applicable" for those two checks; the platform rescans them by itself, a few at a time while the scanner is idle (a failed rescan is retried after an hour, three times at most). The Controls page (/posture/controls) shows each control's full text, how many resources it checked and how many fail; dcloud posture controls list gives the same.

The last control is about Security Posture itself: if a check cannot run, you see it, and nothing that check found before is marked fixed until it runs again.

Findings

A finding is one failing check on one resource (and, where it matters, one rule: a port, an image digest, a policy). Its page shows why it matters, the steps to fix it, a link to the resource, the evidence the check saw, its history, and who changed it:

  • when a change made the check fail, the person (or key, or role) and the time of that change, from the audit log;
  • when a scheduled check found it, whoever last created or changed the resource, also from the audit log ("last changed by");
  • only when the audit log holds no change to the resource (it was made before the log started, or outside the API) does the page say a scheduled check found it.

The change that fixed a finding is recorded the same way.

Fix it

Where one API call fixes a finding, its page has a Fix it button. It first shows the exact call (method, path and JSON body), what changes for the people using the resource, and the permission it needs; nothing happens until you confirm. The call runs as you, through the same API and permission check as any other: it appears in the audit log under your name, and the finding resolves on the re-check a few seconds later. Calls offered today: remove a world-open security group rule, block public access on a bucket, require TLS, set 7 days of backups or turn on deletion protection on a database, deactivate an unused access key, keep the audit log 400 days, add a protection rule to an unprotected default branch, turn on DNSSEC and turn on secret push protection. Copy the call to run it from a script or the CLI instead. Everything else links to the page that fixes it: changes that need a decision (which certificate, which CIDR to allow, a new image) stay yours.

On every resource

Instances, security groups, buckets, databases, Git repositories and registry repositories show their findings where you already look:

  • on the resource's own page, a badge next to its name: 2 findings in the colour of the worst, or Checks pass;
  • in their lists, a Findings column with the count, sortable by severity.

Both link to the resource's open findings. They show counts only, so anyone with posture:GetSummary (part of ReadOnlyAccess) sees them; the findings themselves stay an audited read. GET /posture/api/resource-findings?type=ec2:instance returns the counts for every resource of a type.

  • A finding is resolved only when the same check runs again on the same resource and passes, or when the resource is deleted. A check that fails to run never resolves anything.
  • A resolved finding that fails again is reopened (the count is kept).
  • Evidence never contains a secret value: names, ids, flags, ports, CVE ids and file paths only.

Findings are an attack map. Reading them (posture:ListFindings, posture:GetFinding) is a sensitive action: every allowed read is recorded in the audit log, and the ReadOnly managed policies leave it out. The score, the counts and the control catalog (posture:GetSummary, posture:ListControls) are ordinary reads anyone with ReadOnlyAccess has. Do not paste findings into shared channels, tickets or CI logs.

Accepting a risk

Some findings are intended: a bucket that hosts a public website, a bastion host. Accept risk on the finding records an owner, a reason and an expiry:

  • at most 90 days for critical and high findings, 365 days for medium and low;
  • accepting a critical finding needs posture:AcceptCriticalFinding in addition to posture:AcceptFinding, so you can let more people accept the everyday case;
  • the acceptance lapses when it expires, when the finding becomes more severe, or when a vulnerability appears that it did not list when it was accepted. The finding is then open again.

An accepted finding keeps being checked. It counts as passing in the score, but never as passing in a framework view: an auditor sees an accepted gap as a gap.

Score

The score (0–100) is the severity-weighted share of passing checks: a critical check weighs 10, high 5, medium 2, low 1. Next to it: how many controls pass on every resource. The overview shows the score for the last 90 days.

ISO 27001 and BSI C5

Frameworks → ISO 27001 and BSI C5 list every control of the framework with:

  • Status: failing (an open high or critical finding), accepted risk, passing, not verified (a check has not run yet) or not covered;
  • Covered by the platform: whether your resources on Deutsche Cloud can show it at all. Controls about people, processes and premises (most of ISO 27001 A 5–7) are part of your ISMS, not of your cloud resources; the platform's own side is described in the compliance documents.

The C5 criterion ids follow the published C5:2020 catalogue; check them against the official text before quoting them to an auditor.

Reports for an auditor

GET /posture/api/reports?framework=iso27001 (or c5; dcloud posture reports export) returns a signed statement: what was checked (control ids and versions), when, with what result, the open findings, the accepted risks with their reasons and expiry, the evidence snapshot it was built from and your audit log's chain head at that moment. It is signed with Ed25519 over the canonical JSON of the statement; verify it with the public key in the response or from GET /posture/api/report-keys. Exporting is a sensitive action and is audited.

As a PDF: Download PDF on a framework's page, or GET /posture/api/reports/pdf?framework= (dcloud posture reports export-pdf): the same statement laid out for print — every control of the framework with its status and coverage, what was checked and when, the open findings worst first, the accepted risks, and the evidence and signature. The statement and its signature are attached to the PDF (statement.json, signature.json), so an auditor can verify it from the file alone. Every page is marked confidential.

Every night Security Posture also keeps an evidence snapshot of every result, hashed and chained to the previous day's, so "what did you check on day X" has a stored answer.

Alerts: mail, webhooks, alarms

Events. Every change of a finding is an event on your organization's default event bus, source ews.posture:

Detail type When
Posture Finding Opened A check started failing on a resource, or a fixed one failed again (reopened: true)
Posture Finding Resolved It passed again (reason: passed) or the resource is gone (out-of-scope)
Posture Acceptance Lapsed An accepted risk expired, got more severe or lists a new vulnerability

detail names the finding, control, severity, resource name and rule, and the finding's page (findingUrl). It never carries the evidence. A rule with an HTTP target is a webhook; a queue, function or RabbitMQ target starts your own handling. For example, every new critical finding:

{ "source": ["ews.posture"], "detail-type": ["Posture Finding Opened"], "detail": { "severity": ["critical"] } }

Metrics and alarms. Every evaluation writes metrics to the EWS/Posture namespace: OpenFindings (dimension Severity: critical, high, medium, low) and Score (0–100). A Monitoring alarm on them sends mail and calls webhooks like any other alarm: for example OpenFindings, Severity=critical, maximum > 0 for 1 period, with your security team's address.

API, CLI and MCP

What API CLI
Score and counts GET /posture/api/summary (?project= for a project's resources, ?resource= for one resource) dcloud posture summary get
Findings GET /posture/api/findings?state=open&severity=critical,high dcloud posture findings list
One finding GET /posture/api/findings/{id} dcloud posture findings get
Accept / revoke POST /posture/api/findings/{id}/acceptance, …/acceptance/revoke dcloud posture findings accept, revoke-acceptance
Controls GET /posture/api/controls dcloud posture controls list
Frameworks GET /posture/api/frameworks/{iso27001,c5} dcloud posture frameworks get
Evaluate now POST /posture/api/evaluations dcloud posture evaluations start
Signed report GET /posture/api/reports?framework= dcloud posture reports export
Report as PDF GET /posture/api/reports/pdf?framework= dcloud posture reports export-pdf
Per resource GET /posture/api/resource-findings?type=ec2:instance dcloud posture summary resources

The assistant and MCP clients have the same operations: ask "what is wrong with my security right now" and they read the summary and the open critical and high findings.

Retention

Evaluations, evidence snapshots, scores and resolved findings are kept for 15 months (a full ISO 27001 cycle or a 12-month C5 Type 2 period, plus time for the auditor); ended acceptances for 36 months. Open findings are kept as long as they are open.

Not yet

  • A weekly summary by e-mail (today: an alarm on the metrics, or an event rule).
  • Scans from the outside (open ports, TLS, web exposures) of your public addresses, opt-in.
  • Connecting outside sources (GitHub, another cloud, your identity provider).
  • More controls: volume snapshots, single-node production databases, health checks on public endpoints, environment tags.
Did this page help you?