Configuration, not code
A cloud security posture assessment reads how your cloud accounts are set up and compares it with known-good settings: who can sign in and how, what can be reached from the internet, what is encrypted, and what is logged. Each gap it finds is a misconfiguration, a security risk, or a compliance gap, and often all three.
It answers a different question from monitoring. Security monitoring asks what is happening right now; a posture assessment asks what could happen because of how things are set up. A public storage bucket, a root user without multi-factor sign-in, or a database with an address on the internet needs no clever exploit. It only needs to be found.
What Jutsu checks
Each connected account gets the checks for its provider. They run every day, and you can run them on demand; a new connection is picked up the same day.
- AWS: 37 checks, from public buckets and root-user multi-factor sign-in to CloudTrail in every region, GuardDuty, AWS Config, VPC flow logs, IMDSv2, encrypted disks and databases, public snapshots, and access-key age. Failed Security Hub controls come in as findings too.
- Google Cloud: 20 checks, including public buckets and projects, firewall rules open to the internet, OS Login and Shielded VM, Cloud SQL public addresses, encryption, and backups, downloadable service-account keys, broad Owner and Editor roles, and data-access audit logs. Security Command Center findings come in as well.
- Microsoft Azure: 35 CIS-based checks, including Remote Desktop and SSH open to the internet, SQL auditing, firewalls, and TLS, anonymous storage access, Defender for Cloud plans, activity log export and alerts, and HTTPS on web apps. Defender for Cloud recommendations come in as findings.
- The platforms you build on: GitHub (13 checks, such as branch protection, secret scanning, push protection, Dependabot, and required two-factor authentication), Google Workspace (6, such as 2-Step Verification, super admins, and dormant accounts), Cloudflare (13), Vercel (9), Railway (10), and Supabase (8, including row-level security and point-in-time recovery).
Read-only, with nothing to install
The assessment is agentless. Jutsu reads configuration through roles that can look but not change: an audit role with AWS's SecurityAudit and ViewOnlyAccess managed policies, created from a CloudFormation template, with no access keys; the Viewer and Security Reviewer roles in Google Cloud; and Reader, plus Security Reader for Defender for Cloud, in Azure. GitHub connects through a read-only GitHub App and Google Workspace through read-only scopes.
The categories of findings
The checks fall into a few groups. Each finding is rated low, medium, high, or critical.
- Internet exposure: public buckets and projects, databases with a public address, and firewall rules that open SSH, Remote Desktop, or every port to the internet.
- Identity and access: multi-factor sign-in for the root user and console users, root access keys, old or unused access keys, downloadable service-account keys, and broad owner roles.
- Data protection: encryption of buckets, disks, and databases, encrypted connections, backups, and versions kept of stored files.
- Logging and detection: CloudTrail in every region, data-access audit logs, the Azure activity log, GuardDuty, and Defender for Cloud, so you would see an attack in progress.
- Code and delivery: branch protection, required reviews, secret scanning, push protection, and Dependabot on GitHub.
From finding to fix
A list of problems is only useful if it says what to do next. Each finding in Jutsu comes with the next step.
- Fix steps: the console steps, the exact commands with your resource's real name filled in, a command to confirm the fix, and the permissions you need to make it.
- Grouped by check: a check that fails on ten buckets is one row on the Risks page, not ten, and you can hand the whole group to one owner.
- Check again: after a fix, ask Jutsu to run the check again. The risk closes once it passes.
- Fix with your AI agent: open a finding in Claude Code, Claude Desktop, or Cursor with a prompt that links to it. Nothing is sent until you press Enter, and the Jutsu MCP server gives your agent the fix brief.
The posture report
The cloud account security report covers one AWS, Google Cloud, or Azure account. It opens with an executive summary and the top risks, then lists the findings by check with their fixes, what Security Hub, Security Command Center, or Defender for Cloud raised, the last 30 days of alert activity, the checks that passed, the findings you accepted, and an appendix of failing resources. It exports to PDF, for your team, your board, or a customer's security review.
Your first report is free. When you sign up, Jutsu runs the AWS, Google Cloud, Google Workspace, and GitHub checks once, on any plan, and puts the results in a Posture Report you can save as a PDF. Daily checks and the cloud account security report come with every paid plan.
What to fix first
Severity tells you how bad a finding is; reachability tells you how soon it matters. A sensible order for a small team:
- Anything open to the internet: public storage, databases with a public address, and SSH or Remote Desktop open to everyone.
- The keys to the account: multi-factor sign-in on the root user and every console user, and no root access keys.
- Logging: CloudTrail, audit logs, and the activity log switched on everywhere, so the next problem leaves a trace.
- Data protection and hardening: encryption, backups, and the medium and low findings, worked through as a backlog.
Posture, compliance, and where this stops
The same daily results feed the evidence tasks in Jutsu's Compliance module, for SOC 2, ISO 27001, and 11 more frameworks. SOC 2 for startups explains how. The report itself lists checks, not framework controls.
A posture assessment reads settings. It doesn't watch live activity (that is what monitoring is for), it doesn't scan your application code, and it isn't a penetration test: nothing tries to break in. It also sees only the accounts you connect.
