CSO Online opinion says machine identity breaches need own playbooks
CSO Online says in an opinion piece that SOC playbooks built for people fail for machine identities, so review who owns service accounts and tokens.
- Published
- Ecosystem
- CloudEnterprise softwareCI/CD
- Exploitation
- No, the source is an opinion article about incident response practice and does not report a new exploited flaw
What happened
According to CSO Online, contributor Ashish Mishra published an opinion article on 2026-11-05 saying that incident response playbooks written for human accounts do not work for machine identities. CSO Online The article is labelled Opinion. It does not report a new breach, a new vulnerability or a CVE, and no patch or fixed version is involved.
The author says a human playbook typically starts by asking the user whether a suspicious sign-in was really them, and that a service account cannot answer. He cites CyberArk research that he says puts machine identities at roughly 82 for every human identity, and says very few of them sit behind MFA. These are the author’s statements, and this advisory has not verified the CyberArk figure.
The author describes three earlier incidents as examples. These are his summaries, and this advisory has not checked them against primary sources:
- Cloudflare’s Thanksgiving 2023 incident: the author says attackers used one access token and three service account credentials taken during the earlier Okta support breach. He says Cloudflare had rotated thousands of credentials but missed those four because the team wrongly believed nothing was using them.
- Microsoft and Midnight Blizzard: the author says a password spray hit a legacy test tenant account with no MFA. He says an old test OAuth app there still had elevated access to the corporate environment and was used to reach corporate email.
- Salesloft Drift, August 2025: the author says attackers obtained OAuth tokens used by the Drift chatbot to talk to Salesforce, and for roughly ten days pulled records from customer Salesforce instances. He says Google’s threat intelligence team concluded the attackers were mostly after credentials such as AWS access keys and passwords.
The author’s conclusion is that none of the three needed a clever attacker at the front door, because a credential was issued for a piece of work and left alone after it ended.
Am I affected?
Nothing in the source names an affected product or version. According to the author, the gap applies to any organization whose incident response playbook assumes it can ask a human to confirm activity. It applies in particular where these things are true:
- Service accounts, API keys or OAuth grants have no recorded owner. The author says the trail often ends at a departed contractor.
- Nobody has documented what would break if a high-privilege non-human identity were revoked.
- Nobody is named in advance as allowed to approve an immediate revoke that may disrupt production.
- Logging for cloud audit and workload identity activity is not turned on. The author notes that logs are only available if someone enabled them.
The source does not say how many organizations have these gaps. It also does not confirm that any specific current attack is under way.
What to do now
The steps below are the author’s recommendations as reported by CSO Online. They are not an official advisory.
- Build an inventory of high-privilege non-human identities. For each one, record the responsible team and one or two sentences on what fails if it is removed.
- Sort identities by blast radius. The author would not put a read-only reporting token on the same response path as a cloud admin service principal.
- For the top tier, name in the playbook the role allowed to authorise an immediate revoke, even if it breaks something. The author says executives should own that call.
- Replace “was this you?” triage with questions the logs can answer. Compare where the credential usually signs in from with where it signs in from now. Compare what it holds with what it has used over recent months. Check for new secrets or newly registered apps created by the identity.
- Write separate triggers for rotation and revocation. The author suggests rotation when you are reasonably confident the attacker has not escalated and the integration cannot be lost, and revocation when there is evidence of an ongoing attack.
- Pre-stage replacement credentials. Have application owners practise generating and storing new ones in their secret management system during a low-traffic week, so the vault path in the runbook is known to be current.
- When scoping, look at the whole lifespan of the credential, not only the period after the alert. Add every secret found in the identity to your rotation list.
- For recovery, consider whether the workload can use short-lived tokens, for example through Workload Identity Federation or just-in-time issuance, instead of static secrets.
- Run a tabletop exercise with application owners and the SOC for your highest-privilege integration, from first alert to recovery. Measure how long it takes to identify who approves the revoke.
The author also says that for SOCI-regulated entities, the 12-hour window for reporting a significant cyber incident to ASD may be running while an approval conversation drags on. Check your own legal obligations, because the source gives no further detail.