Please report privately rather than in a public issue: security@hypersurgery.dev.

Include what the operator did, what an attacker could reach, and how to reproduce it. A deployment you can share the shape of (a redacted NetworkScope, the IAM policies in use) helps more than a stack trace. We aim to acknowledge within three working days and to agree a disclosure date with you once the fix is understood; credit is yours unless you ask otherwise.

Worth knowing when judging an issue's severity:

• Read paths never write. Discovery calls ec2:DescribeVpcs, ec2:DescribeSubnets and ec2:DescribeRouteTables, nothing else, through a read-only role per account.
• Writes are off unless asked for twice: the manager needs --enable-writes, and the account needs a write role (spec.accounts[].aws.writeRoleARN) separate from the discovery role. Tagging an existing resource needs only ec2:CreateTags; creating subnets needs ec2:CreateSubnet and friends. The shipped write role allows these on VPCs, subnets and route tables only.
• Nothing is ever deleted. There is no code path that deletes a VPC, a subnet or a tag in a cloud account. Deleting a SubnetClaim or a ResourceImport leaves AWS untouched.
• Credentials are never logged, and role ARNs are checked against the account ID declared in the scope, so a mistyped ARN reports an error rather than reading another account.
• Google credentials for the Sheet export must be a service account key; other credential configurations are rejected because they can make the Google client execute a local command.

The full picture — assets, identities, each trust boundary with its threats, mitigations and residual risks, and a brief for an external reviewer — is in the threat model.

• The operator mirrors resource tags into cluster objects. Anyone who can read Subnet objects can read those tags — put nothing secret in tags, and scope RBAC accordingly.
• Metrics carry account IDs, region names and resource IDs by design; protect the metrics endpoint as you would any other operational data (the chart serves it over HTTPS with authentication by default).

From 1.0, the latest minor release gets every security fix as a patch release, and the minor release before it gets fixes for vulnerabilities rated high or critical for three months after its successor is released. Releases before 1.0 get no fixes: upgrade. The details, and which Kubernetes versions are supported, are in docs/policy.md.