Glossary
Tamper-Evident
You may not be able to stop someone changing a record. You can make sure the change cannot be hidden.
General definition
Tamper-Evident describes a system in which unauthorised changes leave detectable traces. It is a weaker and more honest promise than tamper-proof, which claims changes are impossible; almost nothing is tamper-proof against an administrator with enough access, so security engineering settles for making tampering visible, attributable and therefore deterred. The physical analogy is the seal on a medicine bottle: it does not stop anyone opening it, it guarantees you will know.
Digital mechanisms, roughly in order of strength:
- Hash chains: each log entry includes the cryptographic hash (SHA-256) of the previous entry, so editing or removing any entry breaks every hash after it; periodically publishing the latest hash elsewhere anchors the chain
- Digital signatures over each record or batch by a key the log administrators do not hold, so a rewritten record cannot be re-signed
- Merkle trees: entries hashed into a tree whose root commits to everything below it, allowing efficient proofs that an entry is included and that the log only ever grew; Certificate Transparency (RFC 9162) is the best-known deployment
- Write-once storage: WORM storage and object-lock features in cloud buckets refuse overwrites and deletes for a retention period, enforced by the storage layer rather than the application
- Independent copies: shipping logs in near real time to a system with different administrators, so tampering at the source shows up by comparison
Regulation turns this from good practice into a requirement for communication records. SEC Rule 17a-4(f) requires broker-dealer records to be kept in a non-rewriteable, non-erasable format or with an audit trail that reconstructs any change; the HIPAA Security Rule includes an integrity standard (45 CFR 164.312(c)) requiring mechanisms to confirm that electronic protected health information has not been improperly altered or destroyed; GDPR’s accountability principle expects demonstrable records of processing. A chat message archive that a database administrator could quietly edit satisfies none of these. Tamper evidence is also the foundation for non-repudiation: a signature or timestamp is only useful if the record it was applied to can be shown to be the record still in storage. Note that encryption does not provide tamper evidence; it hides content, and a ciphertext can still be deleted or replaced unless it is also authenticated and chained.
In the Ethora ecosystem
Ethora’s compliance audit trail is designed around this property. Administrative and access events are recorded server-side and exported as immutable logs to S3 in the deployment’s own account; placing those exports in a bucket with object lock enabled gives them storage-enforced write-once protection for the retention period you set, so an operator cannot rewrite history after the fact, and the audit log stays separate from the application database it describes. Message history lives in a read-only server-side archive database with configurable retention, which keeps the record of what was said apart from the live system that could change it.
On a dedicated or self-hosted deployment the bucket, its lock settings and the people who hold its credentials are yours, which is what an auditor will ask about. The Trust and Safety system adds a related control: when sensitive data is removed from a room immediately or after a delay, that action is reported to your systems via message bus, API or email, so a removal is a recorded event rather than a silent edit. Optional end-to-end encrypted rooms, enabled per app, keep the metadata trail but not readable content, which is why they stay off by default for archived deployments.