airBaltic Breach: A Vendor Secret Rotation Runbook

Another week, another supplier breach email. In reports published on November 15, 2026, airBaltic said that someone got unauthorised access to passenger booking data held by an external IT service provider. The usual reaction to news like this is rotating everything in a hurry, or rotating nothing because "it was the vendor, not us". Both are wrong. This post turns the incident into a runbook you can rerun the next time a supplier notification lands.
What airBaltic Disclosed
airBaltic published the notice in its newsroom. According to Aviacionline and LETA via inbox.lv:
- About 618,000 booking records (PNRs) are affected, from passengers who checked in online between 18 July and 12 November 2026.
- The exposed fields are names, booking codes and flight details. Payment data and travel documents were outside the scope of the incident.
- airBaltic says its own information systems were not hacked. The incident happened in the infrastructure of a third-party IT service provider, which the airline has not named.
- For about 11,800 bookings on upcoming flights, online check-in and booking management were disabled, and those passengers can check in at the airport free of charge.
- Passengers are told to watch for phishing emails, SMS and calls about their booking, especially offers of refunds or compensation.
Note what is missing: no passwords, API keys or SSH keys are reported as exposed. For a passenger, the right response is phishing awareness, not a password reset. For an operations team, the lesson is about the supplier link itself.
The Pattern Behind Supplier Breaches
Shortly before, a separate incident showed how these links usually break. The Record reported that several Asian airlines disclosed breaches after Amadeus said it had become aware of "a vulnerability affecting a travel booking display environment." Cathay Pacific said Amadeus told it the threat actors used "login credentials assigned to its system account." Nothing public ties this to the airBaltic incident, but the shape is common: a long-lived credential that a supplier holds, or that you issued to a supplier, becomes the door.
Every integration creates such a credential: a payment webhook secret, a backup vendor's SSH key in authorized_keys, cloud keys in your CI provider. When that vendor sends a breach notice, these are the secrets in question, and they are rarely written down in one place.
Step 1: Triage the Notice
Do not start by rotating. Start by asking what the vendor held of yours. The notice rarely says it, so your inventory has to.
| What the vendor held | Example | Action | Deadline |
|---|---|---|---|
| Credentials you issued to it | IAM key, SSH key, API token for your systems | Rotate and revoke, check last use | Same day |
| Credentials it issued to you | Its API key in your config | Regenerate in its portal, redeploy | Same day once the vendor says the new keys are safe |
| Shared secrets | Webhook HMAC secret, SFTP password | Rotate on both sides together | Same day |
| Your users' personal data only | Names, bookings, emails | No secret rotation; warn users about phishing, harden support processes | Within your notification duties |
| Nothing of yours | You only use a product it sells | Record the review and close | Next working day |
NIST SP 800-63B says verifiers "SHALL NOT require subscribers to change passwords periodically" but "SHALL force a change if there is evidence that the authenticator has been compromised." A supplier breach is that evidence for every secret the supplier could reach, and no evidence for the rest. Rotate on events, not on panic.
Step 2: Keep a Vendor Secret Inventory
The triage table only works if each third-party secret has a row somewhere. Keep the inventory in the same repo or secret manager as the secrets, not in a spreadsheet nobody updates. For each secret, record:
- Vendor and direction: who holds it, and who issued it
- Where it lives: secret manager path, CI variable name, host and file
- Consumers: every service, job and script that reads it
- Scope: what it can read and write, in one sentence
- Rotation method: script, portal click path, or vendor ticket
- Owner: a team, not a person who might leave
- Last rotated and last used
The "consumers" column prevents outages. The "rotation method" column turns a scramble into a routine change.
Step 3: Rotate Without an Outage
The safe pattern is overlap: create the new secret, move consumers, disable the old one, watch, then delete. The AWS IAM procedure is the reference example:
# 1. Create a second key while the first is still active
aws iam create-access-key --user-name vendor-backup-sync
# 2. Update every consumer to the new key, then check the old key
aws iam get-access-key-last-used --access-key-id AKIAOLDKEYEXAMPLE
# 3. Deactivate first, do not delete yet
aws iam update-access-key --user-name vendor-backup-sync \
--access-key-id AKIAOLDKEYEXAMPLE --status Inactive
# 4. After a quiet period, delete
aws iam delete-access-key --user-name vendor-backup-sync \
--access-key-id AKIAOLDKEYEXAMPLE
AWS recommends that you do not delete the old key right away, even if it shows no recent use. Deactivation is reversible. If a forgotten cron job breaks, you turn the key back on, fix the consumer, and try again.
Use the same overlap for webhook secrets (where the provider allows two active secrets during a switch), database passwords (create a second user, then drop the first) and TLS client certificates.
Step 4: Revoke, Do Not Just Replace
Rotation adds a new secret. Revocation removes the old one, and that is the step attackers care about.
- SSH keys. List what is actually trusted on each host with
ssh-keygen -lf ~/.ssh/authorized_keysand compare fingerprints with your inventory. On GitHub, the SSH key review page says to delete keys you do not recognize, because "an unidentified public key may indicate a possible security concern." - Tokens. GitHub recommends an expiration on every personal access token and deleting tokens that are no longer needed. Watch the trap: deleting a token that created a deploy key also deletes the deploy key.
- Dynamic secrets. If vendors get credentials from Vault, revoke a whole path at once. The Vault lease docs show
vault lease revoke -prefix aws/, which invalidates the secrets immediately and deletes the AWS keys. The KV backend for static secrets does not issue leases, so this does not help them.
Short-lived credentials make the next notice cheaper. If you are still handing vendors static SSH keys, look at SSH certificates with step-ca, where access expires on its own.
Step 5: Watch for Use of Old Credentials
After revocation, an old secret should never appear again. Alert on it:
- Authentication failures with the old key ID or fingerprint in CloudTrail or
auth.log - Calls from the vendor's IP ranges outside its normal schedule
- New keys, users or tokens created by the vendor's account during the exposure window
A failed login with a revoked credential is a strong signal that someone else had a copy. Open an incident for it, do not file it as noise.
Common Failure Modes
- One credential, many vendors. A single "integrations" API key shared by three suppliers means one breach forces three migrations. Issue one credential per vendor per purpose.
- Secrets in tickets. The original key went to the vendor by email or in a support ticket. Rotating it in your secret manager does not remove that copy, and the new one must not travel the same way.
- Forgotten consumers. The key is rotated, the main service works, and a weekly report job fails four days later. The inventory's consumer column and the deactivate-first step catch this.
- Vendor-side rotation lag. You rotate a secret the vendor issued, but its own systems are still compromised. Wait for its all-clear before you rely on the new value, and keep the scope of that credential small in the meantime.
- Your own tools hold vendor secrets too. CI systems, monitoring and AI agents store credentials for everything they touch. They belong in the inventory, which is the point of the Keyv npm compromise: CI often holds the keys that matter most.
Where an Operations Agent Helps
The runbook above is mostly lookups, comparisons and tickets, which is work an agent can drive while humans approve the risky changes. In Akmatori, operators write runbooks that the agent searches before it calls any infrastructure tool, as the architecture docs describe. A vendor-notice runbook can tell it to read the inventory, list affected secrets, and check which hosts still trust an old key.
The connectors keep that safe by default. SSH hosts are read-only unless you allow write commands, and you can allow a single diagnostic such as ssh-keygen -lf *. The Jira connector can search issues, and it opens remediation tickets only after you turn on Allow write operations. The agent never sees connector credentials, because the MCP Gateway reads them when it runs a tool. Those connector credentials are still vendor-grade secrets, so give them a row in your inventory too.
Conclusion
The airBaltic incident exposed passenger data, not keys, but it is a good reminder of how much trust runs through suppliers. Triage the notice against an inventory, rotate with overlap, revoke the old secret, and watch for it to come back. Run it once as a drill, and the next breach email becomes a checklist instead of a day of guessing.
If your team wants AI-assisted incident workflows for breach response and secret hygiene, Akmatori helps SRE teams investigate alerts, coordinate response, and automate safe infrastructure actions. Powered by Gcore for global infrastructure reliability.
