THE RECORD · SEP 17 TO OCT 2, 2026
Everyone demos the fix. This is the other half: every change Tamnoon’s engine would not make, what it read first, why it stopped, and what would change its mind. Identifiers removed.
106
alerts it proved safe in the same period, in four changes
20
three changes the engine declined and seventeen it would not make yet
6
fields on every receipt, the same for a yes and a no
RECEIPT NO. 19
The engine read the route table and said no, in writing, with the one fact that turns it into a yes.
Same rule, same day: 23 other subnets came back SAFE.
Intent is not evidence.
No. 19
Finding
Every new instance in this subnet gets a public IP.
What we read first
The route table: one way out, an Internet Gateway. No NAT, no endpoints, no VPN. Three comments: “good to go”.
Verdict
RISKY
Reason
Without a public IP, nothing new in this subnet can reach the internet. “Good to go” never said that was acceptable.
What would change it
A NAT gateway, or the owner saying outbound is not needed.
How to do it by hand
Add the NAT and its route, flip the setting, launch one test instance, rescan.
Finding
A VPC subnet assigns a public IP address to every instance at launch.
What we read first
The subnet’s route table: one route out, through an Internet Gateway. NAT gateways, VPC endpoints, VPN and Transit Gateway attachments in the VPC: none covers this subnet. Three comments on the alert: “confirmed nothing targeting”, “good to go”, “confirmed. not needed”.
Reason
An Internet Gateway cannot route traffic for an instance without a public IP. Turn the setting off and every new instance launched into this subnet loses outbound access, with no alternative path. The comments state intent; none of them says that losing outbound access is acceptable.
What would change it
A NAT gateway or VPC endpoints covering the subnet’s outbound needs, or the owner stating in the record that instances in this subnet need no outbound access. “Good to go” is not that statement.
How to do it by hand
Add a NAT gateway in a public subnet and a default route to it from this subnet; then set MapPublicIpOnLaunch to false; launch one test instance without a public IP and confirm it reaches the internet; then let the rescan confirm.
Before it changes anything, the engine reads who uses the resource, what depends on it, who owns it and what changed. Then it gives one of three verdicts, and every verdict carries the same six fields.
FINDING · WHAT WE READ FIRST · VERDICT · REASON · WHAT WOULD CHANGE IT · HOW TO DO IT BY HAND
SAFE
A proven route, the rollback ready before anything moves, run through your own change process.
RISKY
Declined in writing, with the evidence attached. Nothing runs.
AWAITING DATA
Says exactly what is missing, and asks the person who has it.
Three changes the engine declined and seventeen it would not make yet, grouped by why it stopped. Open any receipt to read all six fields.
No. 19
Finding
Every new instance in this subnet gets a public IP.
What we read first
The route table: one way out, an Internet Gateway. No NAT, no endpoints, no VPN. Three comments: “good to go”.
Verdict
RISKY
Reason
Without a public IP, nothing new in this subnet can reach the internet. “Good to go” never said that was acceptable.
What would change it
A NAT gateway, or the owner saying outbound is not needed.
How to do it by hand
Add the NAT and its route, flip the setting, launch one test instance, rescan.
Finding
A VPC subnet assigns a public IP address to every instance at launch.
What we read first
The subnet’s route table: one route out, through an Internet Gateway. NAT gateways, VPC endpoints, VPN and Transit Gateway attachments in the VPC: none covers this subnet. Three comments on the alert: “confirmed nothing targeting”, “good to go”, “confirmed. not needed”.
Verdict
RISKY
Reason
An Internet Gateway cannot route traffic for an instance without a public IP. Turn the setting off and every new instance launched into this subnet loses outbound access, with no alternative path. The comments state intent; none of them says that losing outbound access is acceptable.
What would change it
A NAT gateway or VPC endpoints covering the subnet’s outbound needs, or the owner stating in the record that instances in this subnet need no outbound access. “Good to go” is not that statement.
How to do it by hand
Add a NAT gateway in a public subnet and a default route to it from this subnet; then set MapPublicIpOnLaunch to false; launch one test instance without a public IP and confirm it reaches the internet; then let the rescan confirm.
No. 20
Finding
A queue and its dead-letter queue, encrypted with the default key.
What we read first
No customer key exists. Three live roles, none with key permissions. Both queues managed by stacks.
Verdict
RISKY
Reason
Three changes in order, on live traffic. One out of order and messages fail.
What would change it
Key created, roles granted, templates updated. Then it is one change.
How to do it by hand
Key, grants, both templates, one deploy window, confirm messages flow, rescan.
Finding
A queue and its dead-letter queue are encrypted with the provider’s default key, not a customer-managed one.
What we read first
The encryption setting on both queues. Every key in the account: no suitable customer-managed key exists. The three roles that touch the queues, two container producers and one serverless consumer: none holds key permissions. The stacks that manage both queues: a console change would drift them.
Verdict
RISKY
Reason
Nothing is unknown. The change is still three changes in order, a new key, permissions on three live roles, two stack updates, across producers and a consumer that are running now. One step out of order and messages fail. The engine does not run a multi-step change on live traffic on its own.
What would change it
The key created, the three roles granted, both stack templates updated. What remains is one change, and that one can be proven safe.
How to do it by hand
Create the customer-managed key with a key policy for the three roles; grant each role permission to generate data keys and decrypt; update both stack templates to reference the key; deploy the two stacks in one change window; confirm messages flow on the main queue and the dead-letter queue; then let the rescan confirm.
No. 21
Finding
A security group on a load balancer accepts port 80 from anywhere.
What we read first
Seven days of flow logs: every accepted port 80 flow came from inside the VPC. Zero from the internet. The load balancer is internal.
Verdict
RISKY
Reason
Nothing is unknown. The change is a rule edit on a security group carrying live traffic.
What would change it
A human on the change. The engine does not edit a live traffic rule on its own.
How to do it by hand
Scope the rule to the gateway’s source group, watch the flows, then rescan.
Finding
A security group on a load balancer permits TCP port 80 from any address on the internet.
What we read first
Seven days of VPC flow logs for the rule: every accepted port 80 flow came from inside the VPC, the API gateway’s private link interfaces, serverless function interfaces and cluster pods. Zero flows from external addresses. The load balancer is internal, in private subnets, with no route from an Internet Gateway. Scoping the rule to the gateway’s source group, or to the specific link addresses, would not disrupt any observed traffic.
Verdict
RISKY
Reason
Nothing is unknown; the blast radius is bounded and evidenced. The change is still a rule edit on a security group carrying live traffic through a load balancer that is serving now. The engine does not edit a live traffic rule on its own.
What would change it
A person on the change. The evidence is already complete; what the engine will not supply alone is the hand on a live rule.
How to do it by hand
Replace the open rule with one scoped to the gateway’s source security group, or to the private link addresses; watch the flow logs through one full traffic cycle; confirm the gateway still reaches the load balancer; then let the rescan confirm.
No. 17
Finding
A bucket policy lets everyone read.
What we read first
The bucket hosts the company logo, embedded in emails and pages we do not control.
Verdict
AWAITING DATA
Reason
Close it and the logo breaks everywhere it is embedded.
What would change it
Where the logo is served from, and which objects must stay public.
How to do it by hand
Put a CDN in front, move the references, restrict the policy, then remove the grant.
Finding
A bucket policy grants read to everyone.
What we read first
The bucket hosts the company logo, and public access to it is intentional. Whether the logo is referenced directly by bucket URL from websites, emails, documentation or external applications; whether a CDN already fronts the bucket; whether other objects in the bucket are also meant to be public.
Verdict
AWAITING DATA
Reason
Removing the global read permission outright would break the logo everywhere it is embedded. The right change is a scoped delivery mechanism, and its shape depends on how the logo is consumed today.
What would change it
The delivery architecture, and the list of objects that must stay public.
How to do it by hand
Put a CDN distribution with origin access control in front of the bucket; move references to the CDN address; restrict the bucket policy to the CDN principal, or to the specific public prefixes; verify; then remove the global grant.
No. 09
Finding
Two CDN distributions respond without security headers.
What we read first
One headers policy per behavior, and both already use a CORS policy the apps need.
Verdict
AWAITING DATA
Reason
The managed policy would replace CORS and break it. Headers without a CSP leave the gap open.
What would change it
The owner choosing the route, and a CSP the apps can live with.
How to do it by hand
One combined policy, CSP in report-only first, then enforce.
Finding
Two CDN distributions respond without security headers.
What we read first
The CDN allows one response headers policy per cache behavior. Both distributions already use a CORS policy the applications need. Three routes: add headers to the existing policy; attach the provider’s managed security headers policy; create a new combined policy with a tailored CSP.
Verdict
AWAITING DATA
Reason
Attaching the managed policy as-is would replace the CORS policy and break CORS. Adding headers without a CSP leaves the main gap open. The route is a design decision the owner has to make.
What would change it
The owner choosing the route, and a CSP the applications can live with.
How to do it by hand
Build a combined policy (CORS, security headers, CSP in report-only first); attach it to both distributions’ behaviors; watch the CSP reports; enforce.
No. 04
Finding
Rotation is off on a secret fifteen serverless services read.
What we read first
Fifteen roles can read it. The identity provider cannot rotate this client in place.
Verdict
AWAITING DATA
Reason
Live consumers or stale grants? No tested procedure. Rotating blind breaks the platform API.
What would change it
A consumer map from the owners, and one rotation tested on a non-production client.
How to do it by hand
List readers from CloudTrail, retire stale grants, roll consumers over one at a time.
Finding
Automatic rotation is disabled on an actively used secret holding OAuth2 client credentials for fifteen serverless consumers across core platform services.
What we read first
The fifteen roles authorized to read it, and which of them read it recently. The identity provider’s app clients require delete-and-recreate rather than in-place rotation. Internal tooling roles also read the secret on the day of assessment.
Verdict
AWAITING DATA
Reason
The misconfiguration is real. The impact of rotating is not established: whether all fifteen roles are live consumers or stale grants, whether a tested rotation procedure exists for this kind of client, and whether the tooling roles are permanent consumers. Rotating blind could break the services behind the platform’s API.
What would change it
A consumer map confirmed by the owners, and a rotation procedure tested once on a non-production client.
How to do it by hand
Enumerate readers from CloudTrail over 30 days; retire stale grants; create the new client, roll consumers over one at a time, verify, then delete the old client; only then enable scheduled rotation.
No. 12
Finding
Three production queues without a customer-managed key.
What we read first
No scheduled producers found. That does not prove there are none.
Verdict
AWAITING DATA
Reason
Encryption adds a key call per message. An unlisted peak producer could throttle.
What would change it
Owners confirming the peak producers, or a review of the schedulers.
How to do it by hand
Confirm peak, raise the key limit if needed, encrypt one queue and watch, then the rest.
Finding
Three production queues without customer-managed key encryption.
What we read first
The automated check for scheduled or periodic producers returned nothing. That does not prove there are none: batch jobs, cron producers and partner feeds can live outside the discovery scope. Observed throughput may not be peak.
Verdict
AWAITING DATA
Reason
Enabling CMK encryption adds a key-management call to every message. If an undiscovered periodic producer sends far more than the observed rate, throttling could cause transient failures at peak.
What would change it
Queue owners confirming the peak producers, or a review of the schedulers (event rules, state machines, scheduled functions, container cron jobs) that publish to these queues.
How to do it by hand
Review the schedulers; confirm peak volume with owners; request a key-management limit increase if needed; enable encryption on one queue first, watch for throttling, then the other two.
No. 07
Finding
A vulnerable image in the registry, not deployed here.
What we read first
Two other accounts can pull it. What they run is not visible from here.
Verdict
AWAITING DATA
Reason
Removing it could leave vulnerable workloads elsewhere, or break a deploy by digest.
What would change it
Confirmation from the owners of the two accounts.
How to do it by hand
Send both owners the digest. When nothing references it, remove it and tighten the policy.
Finding
A vulnerable image in the registry, not deployed in the account that owns it.
What we read first
The repository policy grants pull access to two other AWS accounts. The image is not running in the originating account. Whether the other accounts have pulled or deployed it is not visible from here.
Verdict
AWAITING DATA
Reason
Removing the image without confirming cross-account use could leave vulnerable workloads running elsewhere, or break deployments that reference it by digest.
What would change it
Confirmation from the owners of the two accounts.
How to do it by hand
Send both account owners the digest; check their running workloads for it; once nothing references it, remove the image and tighten the repository policy.
No. 14
Finding
An IAM role with no recent activity.
What we read first
Every key policy in the account except one, which returned AccessDenied.
Verdict
AWAITING DATA
Reason
If that policy names the role, deleting it leaves a dangling principal a future role could inherit.
What would change it
Reading that one key policy.
How to do it by hand
The key administrator checks the policy, removes the reference, then delete the role.
Finding
An IAM role with no recent activity.
What we read first
Every key policy in the account except one. That one returned AccessDenied during the scan, so whether it references the role is unknown.
Verdict
AWAITING DATA
Reason
If the key policy names this role, deleting the role leaves a dangling principal. A future role created with the same name would inherit access to the encrypted data.
What would change it
Reading that key policy, with elevated permission or from the key’s administrator.
How to do it by hand
Have the key administrator check the policy for the role; remove the reference first; then delete the role.
No. 01
Finding
A production database secret, never rotated.
What we read first
Named for a cluster that no longer exists. Its host points at the live replacement.
Verdict
AWAITING DATA
Reason
If the live cluster has its own secret, this one is a leftover. Rotation on a leftover is noise.
What would change it
Which secret the live cluster’s consumers use.
How to do it by hand
List callers over 90 days, re-point any stragglers, then delete or rotate.
Finding
A production database secret has never been rotated.
What we read first
The secret’s name refers to a database cluster that has been deleted. Its host value points at the live cluster that replaced it. Whether a separate, canonical secret already exists for the live cluster.
Verdict
AWAITING DATA
Reason
If the live cluster has its own secret and its consumers were moved to it, this one is a leftover. Setting up rotation on a redundant secret adds machinery to something that should be deleted.
What would change it
Confirmation of which secret the live cluster’s consumers use. A canonical secret exists: delete this one. It does not: configure rotation here.
How to do it by hand
List this secret’s callers in CloudTrail over 90 days; compare with the live cluster’s secret; re-point any remaining consumer; then delete, or enable rotation with a tested rotation function.
No. 02
Finding
A secret behind a replication and analytics pipeline, never rotated.
What we read first
Daily access, on a schedule. Schedules look the same whether or not anyone reads the output.
Verdict
AWAITING DATA
Reason
Unused: delete the pipeline. Used: rotate without breaking three services at once.
What would change it
A stakeholder confirming the output is consumed.
How to do it by hand
Find the downstream consumers. Unused: disable, then delete. Used: rotate in a maintenance window.
Finding
A secret used by a replication and analytics pipeline (a document database replica, a replication task, ETL jobs) has never been rotated.
What we read first
CloudTrail shows daily access. The pipeline runs on schedules, so daily access looks identical whether or not anyone reads its output. Whether any dashboard, application or person consumes what the jobs produce.
Verdict
AWAITING DATA
Reason
Daily API activity is not evidence of business use. If the pipeline is unused, deleting the secret and the infrastructure behind it is the safer change. If it is used, rotation has to be set up without breaking three services at once.
What would change it
A stakeholder confirming whether the pipeline’s output is consumed.
How to do it by hand
Identify downstream consumers of the ETL output; ask the data owner. Unused: disable the schedules, then delete the secret and the infrastructure, in that order. Used: configure rotation with the pipeline paused in a maintenance window.
No. 03
Finding
A per-tenant secret, never rotated.
What we read first
Last access 47 days ago. The function and its trigger are still live.
Verdict
AWAITING DATA
Reason
Paused and decommissioned look the same for 47 days. The right change is the opposite in each.
What would change it
The owner’s word on the tenant’s status.
How to do it by hand
Decommissioned: disable the trigger, delete. Active: rotate, test once.
Finding
A per-tenant secret has never been rotated.
What we read first
Last access 47 days ago. The function that reads it is still active and its queue trigger is enabled. Thresholds: no access in 90 days counts as orphaned, daily access as active. This secret sits between the two.
Verdict
AWAITING DATA
Reason
A paused tenant and a decommissioned tenant look the same from cloud APIs for 47 days. The right change is the opposite in each case: rotate, or delete.
What would change it
The owner’s answer on the tenant’s status. Decommissioned: delete the secret and disable the trigger. Paused or resuming: configure rotation.
How to do it by hand
Ask the service owner named in the record. Decommissioned: disable the queue trigger, delete the secret, remove the tenant’s configuration. Active: set up rotation and test one rotation before the tenant resumes.
No. 13
Finding
A test queue, a week old, unencrypted and idle.
What we read first
No traffic. The creator is known.
Verdict
AWAITING DATA
Reason
Idle is not abandoned. With consent, it is the right place to try encryption first.
What would change it
The creator’s confirmation.
How to do it by hand
Ask. With consent, encrypt here first and verify the consumer path.
Finding
A queue created about a week earlier for local testing, unencrypted and dormant.
What we read first
Metrics show no traffic. The creator is known. Whether it belongs to an upcoming sprint, a scheduled test or a CI/CD pipeline that will resume.
Verdict
AWAITING DATA
Reason
Dormancy is not abandonment; a developer may be about to use it. With the owner’s consent it is also the right place to try the encryption change before touching production queues.
What would change it
The creator’s confirmation.
How to do it by hand
Ask the creator; with consent, enable CMK encryption here first and verify the consumer path; use the result to plan the production queues.
No. 15
Finding
Two IAM roles with no recent activity.
What we read first
Both are wired into live CI/CD: build projects, pipelines, the stacks they deploy.
Verdict
AWAITING DATA
Reason
Idle and retired look the same from the role. Only the team knows.
What would change it
The platform team confirming the pipeline is retired.
How to do it by hand
Retired: disable the pipeline, then the stacks, then the roles. Needed: keep, note the date.
Finding
Two IAM roles with no recent activity.
What we read first
Both are embedded in CI/CD: active build projects, pipelines and the stacks they deploy. The pipelines have not run recently; the infrastructure is intact.
Verdict
AWAITING DATA
Reason
An idle pipeline and a retired pipeline look the same from the role’s activity. Only the owning team knows whether deployments still depend on it.
What would change it
The platform team confirming the pipeline is retired or moved.
How to do it by hand
Ask the owning team. Retired: disable the pipeline, delete the stacks it manages if they are retired too, then the roles. Still needed: leave the roles and record the review date.
No. 05
Finding
A browser vulnerability reported on a container image.
What we read first
Scan on Push is off, so there is no package inventory for the image.
Verdict
AWAITING DATA
Reason
Whether the browser is in the image, which version, and where it runs: all missing.
What would change it
An image scan and a list of workloads built from it.
How to do it by hand
Enable scanning, scan, list deployments by digest, then rebuild, patch or remove.
Finding
A use-after-free vulnerability in Chrome/Chromium, versions before 153.0.8010.36, reported on a container image in the registry.
What we read first
The repository has Scan on Push disabled, so no package inventory or vulnerability data exists for the image. No automation results are available.
Verdict
AWAITING DATA
Reason
It is not established that the browser is present in the image, which version it is, which running containers are built from it, or how far it is deployed across the clusters. Every factor that decides the impact of a change is missing.
What would change it
An image scan (enable Scan on Push and re-push, or scan on demand) and a list of workloads built from the image.
How to do it by hand
Enable scanning on the repository; pull and scan the image; list deployments by image digest across clusters; then decide: rebuild, patch in place, or remove the browser.
No. 06
Finding
A browser vulnerability on an image whose origin is unknown.
What we read first
Nothing shows how the image was built, or by whom.
Verdict
AWAITING DATA
Reason
Without a pipeline, a rebuild is a manual, error-prone act.
What would change it
The original build method and its owner.
How to do it by hand
Find the author, recover or write a Dockerfile, rebuild under a new tag, roll deployments.
Finding
A browser vulnerability on a container image whose origin is unknown.
What we read first
No automation in the library can determine how the image was built (a snapshot of a configured instance, a one-off build), whether build instructions exist, or who built it.
Verdict
AWAITING DATA
Reason
Without a build pipeline, rebuilding the image is a risky, error-prone manual act. Whether a safe rebuild is possible, or whether patching in place or removing the browser is the right change, depends on how the image was made.
What would change it
The original build method and its owner.
How to do it by hand
Find the image’s author from repository and CI history; recover or write a Dockerfile; rebuild with the patched browser under a new tag; test; roll deployments to the new tag; retire the old digest.
No. 08
Finding
A deployment stack with no notification topic.
What we read first
Actively managed through CI/CD. No evidence of which topic, or who may publish to it.
Verdict
AWAITING DATA
Reason
Adding a target to a live deployment is a change to the deployment itself.
What would change it
The owner naming the topic, and the deploy role’s permission to publish.
How to do it by hand
Create the topic, grant publish, add it in the pipeline’s parameters, deploy through CI/CD.
Finding
A CloudFormation stack has no notification topics configured.
What we read first
The misconfiguration is real: the list is empty. The stack is actively managed through CI/CD. No evidence of which topic should receive the notifications, whether the deployment role may publish to it, or whether any operational process depends on receiving them.
Verdict
AWAITING DATA
Reason
Adding a notification target to an actively deployed stack is a change to the deployment itself, and its impact is not established.
What would change it
The owner naming the topic and confirming the deployment role’s publish permission.
How to do it by hand
Create or choose the topic; grant publish permission to the deployment role; add the topic in the pipeline’s template parameters, not by hand in the console; deploy through CI/CD.
No. 10
Finding
Missing security headers on the marketing website.
What we read first
Hosted on a third-party platform. Nothing in the account serves it.
Verdict
AWAITING DATA
Reason
If the platform controls the headers, there is nothing to change in AWS.
What would change it
Whether the platform allows custom response headers.
How to do it by hand
Set them on the platform, or put a CDN in front. Otherwise close as not actionable, note attached.
Finding
Missing security headers on the company’s marketing website.
What we read first
The site is hosted on a third-party website platform. No CDN distribution, load balancer, API gateway or VPC endpoint in the account serves it.
Verdict
AWAITING DATA
Reason
If the platform controls the response headers, there is nothing to change in AWS and the finding is not actionable here. If a platform-level or DNS-level mechanism to set headers exists, the work is real but outside the cloud account.
What would change it
Confirmation of whether the platform allows custom response headers.
How to do it by hand
Check the platform’s header settings, or put a CDN with header rules in front of it; otherwise close the finding as not actionable in AWS, with this note attached.
No. 11
Finding
Header findings on 20 VPC endpoints.
What we read first
All 20 are provider-managed. There is no place to set a header.
Verdict
AWAITING DATA
Reason
The finding describes a change that cannot be made.
What would change it
A new endpoint to a non-provider service, where header control exists.
How to do it by hand
Record the classification, mark the 20 not actionable, check new endpoints at each review.
Finding
Header injection findings on 20 VPC endpoints.
What we read first
All 20 are provider-managed service endpoints or auto-managed serverless cache endpoints. Gateway endpoints operate at the routing layer with no HTTP interception. Interface endpoints for provider services return responses the provider controls. None connects to a service the company or a third party operates.
Verdict
AWAITING DATA
Reason
There is no place in these endpoints to set a response header. The finding describes a change that cannot be made.
What would change it
A new endpoint connecting to a non-provider service, where header control would exist.
How to do it by hand
Record the classification; mark the 20 as not actionable; add a check for new endpoints to the review so the classification stays current.
No. 16
Finding
Two cross-account roles with no recent activity.
What we read first
Both let security vendors in, one to scan, one to onboard.
Verdict
AWAITING DATA
Reason
Whether a vendor is still needed is a business decision, not a cloud fact.
What would change it
The vendor owner confirming the integration is retired.
How to do it by hand
Retired: remove the role and the trust. Active: keep it, record why it is quiet.
Finding
Two cross-account IAM roles with no recent activity.
What we read first
Both are integration points for third-party security vendors, one for scanning and one for onboarding. Removing them removes the vendor’s ability to scan or manage the account.
Verdict
AWAITING DATA
Reason
Whether a vendor integration is still required is a business and security decision. It cannot be read from cloud configuration.
What would change it
The security team or vendor owner confirming the integration is retired.
How to do it by hand
Confirm with the vendor relationship owner. Retired: remove the role and the vendor’s account from any trust policy. Active: keep it, and record why the role is quiet.
RECEIPT NO. 20
The engine found every fact it needed. The change was still three changes in a row on live producers. It declined and wrote the order down.
Understood is not the same as safe in one step.
No. 20
Finding
A queue and its dead-letter queue, encrypted with the default key.
What we read first
No customer key exists. Three live roles, none with key permissions. Both queues managed by stacks.
Verdict
RISKY
Reason
Three changes in order, on live traffic. One out of order and messages fail.
What would change it
Key created, roles granted, templates updated. Then it is one change.
How to do it by hand
Key, grants, both templates, one deploy window, confirm messages flow, rescan.
Finding
A queue and its dead-letter queue are encrypted with the provider’s default key, not a customer-managed one.
What we read first
The encryption setting on both queues. Every key in the account: no suitable customer-managed key exists. The three roles that touch the queues, two container producers and one serverless consumer: none holds key permissions. The stacks that manage both queues: a console change would drift them.
Reason
Nothing is unknown. The change is still three changes in order, a new key, permissions on three live roles, two stack updates, across producers and a consumer that are running now. One step out of order and messages fail. The engine does not run a multi-step change on live traffic on its own.
What would change it
The key created, the three roles granted, both stack templates updated. What remains is one change, and that one can be proven safe.
How to do it by hand
Create the customer-managed key with a key policy for the three roles; grant each role permission to generate data keys and decrypt; update both stack templates to reference the key; deploy the two stacks in one change window; confirm messages flow on the main queue and the dead-letter queue; then let the rescan confirm.
RECEIPT NO. 18
A refusal count means nothing without the yes beside it. 106 alerts came back SAFE this period, in four changes. One of them:
No. 18
Finding
A pipeline role can write S3 object ACLs.
What we read first
90 days of CloudTrail: zero calls. Both buckets have ACLs disabled, so the call cannot work.
Verdict
SAFE
Reason
Unused, unusable, and nothing in the deploy path needs it.
What would change it
One such call in CloudTrail, or a bucket turning ACLs back on.
How to do it by hand
Back up the policy, remove the one action, re-apply. Rollback: the backup.
Finding
A CI/CD pipeline role’s inline policy grants s3:PutObjectAcl, write access to S3 object ACLs.
What we read first
Ninety days of CloudTrail for the role: zero PutObjectAcl calls. Both target buckets enforce bucket-owner object ownership, which disables ACLs entirely, so any such call would fail with AccessControlListNotSupported. The deploy stage: a CDN and a bucket policy, no object ACLs.
Reason
The permission has not been used, cannot work on these buckets, and nothing in the deploy path depends on it. Removing one action leaves every other permission, and the running pipeline, untouched.
What would change it
A PutObjectAcl call in CloudTrail, or either bucket moving off bucket-owner-enforced ownership.
How to do it by hand
Back up the inline policy to JSON. Remove s3:PutObjectAcl from every statement’s Action list and re-apply the policy. Confirm the action is gone. To roll back, re-apply the backup.
Half of everything ever detected in the cloud is still open: 53% across 14.86 million detections, up from 41% a year earlier. A critical finding now stays open 150 days on average.
The findings were never the hard part. The change is. Nobody approves a change in production that a machine cannot explain, so the remediate buttons stay off and the backlog grows.
So the first thing we automated was restraint. Every vendor can show you a fix. This page is the other half.
Everyone demos the fix. Ask them to demo the decline.
Six questions, word for word. A product with judgment passes them. Use them on everyone, including us.
01
Can it refuse a fix? Show us one refusal, with the evidence attached.
02
Does it give a verdict on findings from every scanner we run, or only its own?
03
Show the full evidence trail for one closed finding: what was read, what was decided, who approved, what the rescan showed.
04
What happens when our engineer overrides it? Where does that no go?
05
Does the fix stay fixed, or does it come back at the next deploy?
06
Can our auditor read every change without a translator?
Copied. Paste them into your next RFP.
HOW THIS RECORD WAS MADE
Period: Sep 17 to Oct 2, 2026. Each receipt is condensed from the engine’s record into the six fields; the verdicts are as the engine issued them. Account IDs, resource names, people and vendor names are removed. In this period the engine issued three RISKY verdicts, all on Sep 30, all in this record. Every other change it did not make it held as AWAITING DATA, with the missing fact named. SAFE means proven safe and ready to run through the change process; it is not counted as closed until the rescan says so. The alert counts are the engine’s own tallies per initiative. The market figures come from Tamnoon’s State of Cloud Remediation 2026 research, 14.86 million detections across hundreds of environments and ten cloud security platforms.
Tamnoon. The self-healing engine for cloud exposures. Tamnoon proves each remediation safe and runs it through your own change process. The rest are declined in writing, with the evidence.
Everyone automates the yes. We mastered the no.
Bring us your ugliest finding.
The one that has been open for most of a year because nobody wants to be the person who touches it. You get a written verdict: what we read, what we decided, why, what would change it, and how to do it by hand if the answer is no.
“I found Tamnoon to be quite compelling from both an innovation standpoint and a productivity standpoint as it helps drive KPI and SLA goal achievement.”
John Remo · SVP, Global Infrastructure & Cybersecurity · Warner Music Group
Watch mode: read-only, 30 minutes, nothing written. You keep the decision record, including the findings Tamnoon declines in writing.
PICK A TIME →