Security

How plainfra keeps your infrastructure safe

Last updated: October 2026 ยท Questions? security@triont.com

plainfra is built on a simple principle: it should not need access to your application data. The AWS role is read-only, the deny policy blocks data contents, and you can remove the role yourself from CloudFormation.

What plainfra can inspect

  • Security groups, route tables, load balancers, and instance metadata
  • Cost Explorer, Budgets, and Savings Plan coverage
  • CloudWatch alarms, backup settings, and health states
  • S3 bucket configuration, versioning, encryption, and public access settings

What plainfra cannot access

  • It cannot create, update, or delete AWS resources, or start a job, other than the CloudWatch Logs Insights queries it reads logs with
  • It cannot run a managed AI service on your behalf, such as text, image or speech analysis
  • It cannot read S3 object contents, S3 table data, or files
  • It cannot read database rows, DynamoDB items, stream records, query results, or queue messages
  • It cannot read source code, including Lambda function packages, chat messages, contact centre recordings and transcripts, or support case attachments
  • It cannot read your end users: user pool and identity pool users, their emails and phone numbers, your messaging audiences, or the email addresses, phone numbers and devices subscribed to your notification topics
  • It cannot read Secrets Manager secret values or SecureString SSM parameters
  • It cannot decrypt with your KMS keys, except so that CloudWatch Logs can return encrypted log groups

1. The IAM role: read-only with explicit denials

When you connect your AWS account, you deploy a single CloudFormation stack that creates an plainfra-ReadOnly IAM role. Configuration and logs are evidence; the contents of your data are not. The role has two parts:

  • Read access (Allow). The AWS managed ReadOnlyAccess policy, so plainfra can read how every service in the account is configured, what it costs, and its CloudWatch logs. It grants no write. It does include a few calls that start a check or a report, or run a managed AI service, and the boundary below denies those. The one exception is starting and stopping a CloudWatch Logs Insights query, which is how plainfra reads logs.
  • A data boundary (Deny always wins). An inline policy, PlainfraBoundary, that overrides the Allow, and anything AWS adds to it later:
    • s3:GetObject, s3:GetObjectVersion, s3express:CreateSession: cannot read file contents
    • dynamodb:GetItem, dynamodb:Query, dynamodb:Scan, and the same reads through DAX, Keyspaces, SimpleDB and OpenSearch: cannot read table rows or documents
    • kinesis:GetRecords, athena:GetQueryResults, redshift-data:GetStatementResult: cannot read stream data or query results
    • codecommit:GetFile, codecommit:GitPull, codecommit:GetDifferences, lambda:GetFunction: cannot read source code, diffs, review comments or function packages
    • cognito-idp:ListUsers, cognito-idp:AdminGetUser, mobiletargeting:GetEndpoint, sns:ListSubscriptions: cannot read your end users, messaging audiences or topic subscribers
    • transcribe:Get*, healthlake:ReadResource, kendra:Query: cannot read transcripts, health records or search results
    • connect:GetContactAttributes, connect:ListRealtimeContactAnalysisSegments, support:DescribeAttachment: cannot read contact centre contacts and transcripts, or support case attachments
    • secretsmanager:GetSecretValue, cognito-idp:DescribeUserPoolClient: cannot read secret values, including app client secrets
    • kms:Decrypt: cannot decrypt anything, so SecureString parameters, objects encrypted with your own keys and Lambda environment variables stay unreadable (AWS decrypts a function's variables with the caller's own kms:Decrypt). ECS task definition values carry no such encryption and can be read as configuration; values that look like credentials are masked before they are stored or shown. The one exception is CloudWatch Logs, which may decrypt a log group encrypted with your own key so that logs stay readable
    • ecr:GetAuthorizationToken, ssm:GetAccessToken, sts:GetSessionToken, codepipeline:GetJobDetails and other token issuers: cannot obtain credentials
    • sqs:ReceiveMessage: cannot read queue messages
    • lambda:InvokeFunction, ssm:SendCommand: cannot execute anything
    • ec2:GetPasswordData: cannot retrieve Windows passwords
    • rds-data:ExecuteStatement: cannot run SQL queries
    • comprehend:Detect*, rekognition:Detect*, polly:SynthesizeSpeech, cloudformation:Detect*, iam:Generate*: cannot run managed AI services, start drift detection or generate reports

Logs are the main place your own data can reach plainfra: application logs carry whatever your applications write to them. Plain Systems Manager parameter values are readable as configuration. If you want log access narrowed to named log groups, we will scope it with you.

plainfra can tell you that an S3 bucket exists, when it was created, its versioning state, and its access control configuration. It cannot read a single byte of object data stored inside it, and the same holds for S3 table buckets and S3 on Outposts.

You can inspect the full CloudFormation template before deploying. We encourage it. Paste it into any AI assistant and ask it to explain exactly what the role can and cannot do.

Prefer to start even smaller? Signup also offers an EC2-only lite role with a single permission, ec2:Describe*. You can upgrade it to the full read-only role at any time by re-deploying the same stack.

Approving this access for your organisation? Read the reviewer's version: Access and revocation, in full, with the commands to verify it yourself.

2. Cross-account trust with ExternalId

The role trust policy restricts assumption to named service roles inside plainfra's own AWS account (121754142582, owned by Triont), not the whole account: two on a standard install, and a third, plainfra-investigation-gateway, once the investigation service is enabled for your account. Every one of them must present a unique ExternalId generated specifically for your account at signup. This prevents the confused deputy problem. Even if an attacker knew plainfra's account ID, they cannot assume your role without your exact ExternalId, and nothing else in plainfra's account is trusted to try. The ExternalId is shown to you in the console when you add an account, and is stored only as part of your encrypted account record. Treat it as a credential. On the AWS path, all AI model calls are made from a second AWS account that Triont owns (section 7). Your role does not trust that account, and the role plainfra uses there can only call AI models and read the subscription check those calls need. That account manages plainfra's own account, so it sits inside plainfra's trust boundary rather than outside it.

3. API keys

Your plainfra API key is shown to you once: at the end of the signup flow and in the welcome email. We store only the SHA-256 hash of your key in DynamoDB. There is no way to recover your plaintext key from our systems. If you lose it, you contact support and we issue a new one.

API keys are prefixed with tt_ to make them identifiable in logs and secret scanners. We recommend treating your API key like a password: store it in a secrets manager, rotate it if you suspect compromise, and never commit it to source control.

4. Data storage and encryption

plainfra stores customer account records, conversation history, generated reports, and infrastructure metadata on AWS infrastructure in Sydney (ap-southeast-2).

  • DynamoDB. Customer account records, subscription state, and API key hashes. Encrypted at rest using AWS-managed keys.
  • S3. Conversation history and generated reports, stored under your customer ID prefix. Encrypted at rest using AWS-managed keys.
  • In transit. All communication between your browser, plainfra APIs, and AWS services uses TLS 1.2 or higher. There is no unencrypted path.

Conversation history is no longer accessible after your plan's session-history period (90 days on Scout, Core and Team; 365 days on Enterprise) and is deleted within 365 days. Reports are kept for 365 days. Discovery snapshots are thinned over time and removed after 12 months. Usage records (token counts and timings, never your question text) are kept for billing. Support-chat history is kept for support quality until you ask us to delete it.

5. Payment security

Payments are handled entirely by Stripe, a PCI DSS Level 1 certified payment processor. plainfra never receives, stores, or processes your card number, CVV, or full card details. No card is taken at signup. When you upgrade, your card details go directly from your browser to Stripe's hosted checkout. Our backend only receives Stripe's confirmation of the subscription.

6. Infrastructure access controls

plainfra runs as serverless AWS Lambda functions and short-lived container tasks, with no servers to patch and no long-lived credentials. Every infrastructure change is applied by CloudFormation from the deployment pipeline, under a dedicated deploy role that the pipeline assumes through GitHub Actions OIDC.

The plainfra Lambda functions themselves run with minimum-privilege execution roles scoped to only the DynamoDB tables, S3 bucket, and AI APIs they need to operate.

7. AI and conversation data

plainfra uses AI inference to answer your questions and generate reports. plainfra uses models from a single model developer. They run inside AWS through its managed model service, or on the developer's own service. On the AWS path, by default, model calls run in whichever AWS region has capacity. They stay within the AWS network, but may be processed outside Australia. Australian primary processing is available as a paid option on request: model calls run in AWS Sydney, and if Sydney is unavailable they run in another AWS region, still within the AWS network.

Data processed by models running inside AWS is never used to train AI models, under AWS's service terms. On the AWS path, all model calls are made from a second AWS account that Triont owns, which holds the AI model access those calls use.

Conversation history is stored in S3 so plainfra can maintain context across sessions. This is separate from AI processing and is subject to the retention periods above.

8. Revoking access

You control the IAM role at all times. To revoke plainfra's access to your AWS account:

  1. Log into your AWS Console
  2. Navigate to CloudFormation > Stacks
  3. Delete the plainfra-ReadOnly stack

The full removal and incident-revocation procedure, including how to invalidate sessions that are already issued, is documented on Access and revocation.

Deleting the stack removes the role within about a minute, and plainfra will no longer be able to query your account. Deleting the role also ends sessions already issued. To cut off access without deleting the role, follow the incident steps on Access and revocation. Your account record and conversation history in plainfra's systems are retained per our Privacy Policy and can be deleted on request.

9. Vulnerability disclosure

If you discover a security vulnerability in plainfra, please report it to security@triont.com. We ask that you:

  • Give us reasonable time to investigate and fix the issue before public disclosure
  • Avoid accessing or modifying data that doesn't belong to you
  • Not perform denial-of-service attacks or disruptive testing

We will acknowledge all reports within 2 business days and keep you informed as we investigate. We do not currently offer a formal bug bounty program, but we genuinely appreciate responsible disclosure and will credit researchers who want to be named.

10. Contact

For security questions or concerns: security@triont.com

For general support: support@plainfra.com