Most threat intelligence programmes are built around malware hashes and IP addresses, while the attacks that actually reach data now run on stolen identities: passwords from stealer logs, session cookies, OAuth tokens and API keys. Mandiant found that at least 79.7 percent of the accounts used in the 2024 Snowflake customer campaign had prior credential exposure, some dating back to infostealer infections in November 2020, which means the intelligence existed years before the theft. This guide shows how to run threat intelligence with identity and non-human identity (NHI) as the subject: what to ask for, which sources produce identity findings, how to match them against your own accounts and secrets, and how to turn a finding into a revocation within hours. It is written for threat intelligence analysts, SOC leads and identity teams. Monitoring criminal markets is covered in the Dark Web Monitoring Guide; this page is about the identity intelligence cycle end to end.
Key takeaways
- Identity is the indicator. For identity attacks the most useful intelligence is not an IP address but a credential, token, account name, OAuth client or service principal that you can match to your own inventory and revoke.
- Non-human identities need their own requirements. API keys, CI tokens, OAuth grants and service accounts leak through code, support tickets, SaaS suppliers and stealer logs, and they rarely have MFA to fall back on.
- Matching beats collecting. A feed is only useful if you can join it to an owner and a system. Without an identity inventory, intelligence becomes a pile of unactionable alerts.
- Speed is measured in revocations. The metric that matters is time from a credible finding to the credential being dead, not the number of indicators ingested.
- Share patterns, never secrets. Share account types, token prefixes, user agents and campaign behaviour under the right TLP label; never forward live credentials outside the team that must revoke them.
Why identity needs its own intelligence focus
Classic indicator feeds age in days: attackers rotate infrastructure faster than block lists update. Identity intelligence ages differently. A stolen password stays valid until someone changes it, a refresh token until someone revokes it, and a static API key until someone rotates it. In the Snowflake campaign some credentials had not been rotated for up to four years. That changes priorities in three ways:
- Old findings matter: a two-year-old stealer log entry for a contractor laptop can still be the key to a data warehouse today. Retroactive matching against historical leaks is worth doing.
- Your suppliers' identities are your exposure: in the 2025 Salesloft Drift incident, the actor tracked as UNC6395 used OAuth tokens that Drift held for its customers' Salesforce integrations, then searched exported data for AWS access keys, passwords and Snowflake tokens. Customer organisations never had a credential stolen from their own systems. See the Salesloft Drift Breach 2025.
- Behaviour outlives indicators: attackers who abuse valid accounts look like users. Intelligence about how they behave (client application names, user agents, VPN providers, the order of their API calls) is more durable than any single IP.
Intelligence requirements for identity and NHI
Write priority intelligence requirements (PIRs) that the identity team agrees to act on. Each needs an owner and a response playbook, or it should not be collected.
| Requirement | Example question | Primary sources | Action owner |
|---|---|---|---|
| Workforce credential exposure | Are any employee or contractor credentials for our domains in stealer logs or combo lists? | Stealer log and credential monitoring services, breach notification services | Identity operations |
| Session and token theft | Are session cookies or tokens for our SSO, mail or SaaS tenants being sold? | Stealer log contents, criminal market monitoring | Identity operations, SOC |
| NHI secret exposure | Are API keys, cloud keys or CI tokens for our accounts public in code, packages, paste sites or container images? | Code host secret scanning, partner alerts, own public scanning | Secrets owner, platform team |
| Third-party identity compromise | Has a SaaS supplier or integration holding tokens to our tenants been breached? | Vendor advisories, ISAC sharing, threat research blogs | Third-party risk, SaaS owners |
| Actor targeting | Which groups are targeting our sector through help desk social engineering, MFA fatigue or OAuth consent? | Government advisories, vendor threat reports, ISACs | Threat intelligence, IAM architecture |
| Identity infrastructure vulnerabilities | Are our identity providers, VPNs or federation services affected by actively exploited flaws? | CISA Known Exploited Vulnerabilities catalogue, vendor advisories | Vulnerability management |
Sources that produce identity findings
Stealer logs and credential markets
Infostealers collect saved browser passwords (T1555.003 Credentials from Web Browsers), session cookies (T1539 Steal Web Session Cookie) and files, and the resulting logs are sold or shared. Mandiant named VIDAR, RISEPRO, REDLINE, RACOON STEALER, LUMMA and METASTEALER as the families behind the Snowflake credentials, and noted that contractor laptops used for personal activity were a common infection source. A single log tells you the infected machine, the URLs credentials were saved for, and often cookies that bypass MFA. Treat a log naming your SSO or admin portals as a live incident, not a statistic. Response steps are in the Infostealer Logs Monitoring and Response Guide.
Code hosts and package registries
Public repositories, gists and packages are the largest source of leaked NHI secrets. GitHub's secret scanning partner programme lets service providers register token formats; GitHub scans public repositories and public npm packages and posts matches to the provider's endpoint, and its documentation tells providers to treat any reported secret as public and compromised. If you issue tokens to customers, joining the programme turns other people's leaks into your intelligence. If you consume tokens, confirm your providers participate and that their revocation alerts reach a monitored mailbox.
Supplier and SaaS advisories
A breach at a SaaS supplier often arrives first as a vendor advisory or threat research post. Map each advisory to the integrations, OAuth grants and service accounts that supplier holds in your tenants; the Gainsight Salesforce Breach 2025, where long-lived OAuth tokens opened customers' CRM data, shows why the map must exist before the advisory lands.
Government and sector advisories
Advisories describe identity tradecraft in detail. CISA's advisory AA23-320A on Scattered Spider (updated 21 November 2023) describes help desk impersonation to reset passwords and MFA, lookalike domains such as "victimname-sso" and "victimname-okta", MFA push fatigue, SIM swapping, and persistence by adding a federated identity provider to the victim's SSO tenant. Each of those is a detection you can build. See the Scattered Spider Threat Actor Profile.
Your own telemetry
Identity provider sign-in logs, SaaS audit logs and cloud API logs are intelligence sources. Failed sign-ins for accounts that do not exist, or for test tenants, often show which leaked lists attackers are working through.
Identity indicators and how to use them
Identity intelligence produces different indicator types from malware intelligence. Each has a different use and lifetime.
| Indicator type | Example | How to use it | ATT&CK reference |
|---|---|---|---|
| Exposed credential | Username and password for an SSO domain from a stealer log | Force reset, revoke sessions, investigate the infected device owner | T1589.001 Credentials; T1078.004 Cloud Accounts |
| Stolen session or cookie | Session cookie for the identity provider or mail | Revoke all sessions for the user, review recent session activity | T1539 Steal Web Session Cookie |
| Application token | OAuth refresh token held by a breached integration | Revoke grants for the app, rotate secrets the app could read | T1528 Steal Application Access Token; T1550.001 Application Access Token |
| Static NHI secret | Cloud access key in a public repository | Disable, rotate, search cloud logs for its use | T1552.001 Credentials In Files |
| Client fingerprint | Unusual client application ID or user agent seen in a campaign | Hunt in sign-in and API logs, alert on reuse | Behavioural, campaign specific |
| Access infrastructure | Commercial VPN or VPS provider used for access and exfiltration | Conditional access risk signal; short-lived block | Behavioural, campaign specific |
Mandiant's Snowflake report is a good model of identity indicators: alongside IP addresses it published client application IDs used by the actor (for example a reconnaissance utility it calls FROSTBITE, and specific JDBC, Go and Python connector versions), and noted that access came mainly through Mullvad and Private Internet Access VPN addresses with exfiltration through ALEXHOST SRL (AS200019) infrastructure. Client names and connector versions in your own data warehouse logs are exactly the kind of signal you can hunt for.
Matching intelligence to your identity inventory
An identity finding is actionable only when it resolves to an account, a secret, an owner and a system. Build the join before you buy the feed.
- Normalise identifiers: map email addresses, user principal names, legacy domains and contractor domains to one person record; map service principals, OAuth client IDs and key IDs to an NHI record with an owner.
- Recognise token formats: structured prefixes make leaked secrets identifiable. GitHub, for example, uses ghp_ for personal access tokens, gho_ for OAuth access tokens, ghu_ for user-to-server, ghs_ for server-to-server and ghr_ for refresh tokens, with a CRC32 checksum in the last six characters so scanners can reject fakes offline. AWS access key IDs begin with AKIA, the string UNC6395 searched for in stolen Salesforce data.
- Match without spreading secrets: compare findings against your inventory using key IDs, fingerprints or salted hashes where possible, so the raw secret is handled by as few people and systems as possible.
- Check liveness safely: decide who may test whether a leaked secret works, against which systems, and how it is logged. For your own keys, checking the key's status in the issuing console is usually safer than trying it.
- Attach context: privilege level, MFA state, last use, and whether the account is human, service or third-party. A finding for an admin with no MFA outranks a hundred finds for disabled accounts.
This is why inventory work and intelligence work are one programme. See the NHI Ownership Guide for building the NHI side of the join, and the Detection Engineering for Non-Human Identities Guide for turning findings into detections.
From finding to action
| Finding | First action (target time) | Follow-up |
|---|---|---|
| Live workforce credential plus cookie in a fresh stealer log | Revoke sessions and reset password (within 1 hour) | Reimage the infected device, review sign-ins since the infection date, check for new MFA methods |
| Old credential (more than 12 months) for an active account | Confirm it is no longer valid; reset if it is (within 24 hours) | Enforce phishing-resistant MFA on the account |
| Cloud key or CI token public in code | Disable and rotate (within 1 hour) | Search cloud and CI logs for use from the exposure time onward |
| Supplier advisory: integration tokens stolen | Revoke the integration's grants and tokens (same day) | Rotate any secrets the integration could read, hunt for its client ID in audit logs |
| Actor advisory naming identity tradecraft | Check detections exist for each technique (within 1 week) | Tabletop with the help desk if social engineering is involved |
Decision guidance: small organisations get most value from free sources (code host secret scanning, provider revocation alerts, government advisories and breach notification services) plus a disciplined join to their identity provider. Commercial stealer log coverage pays off once you have more than a few thousand identities, contractors with unmanaged devices, or a history of credential-based incidents. Large organisations should give identity findings their own queue with service levels owned by identity operations, not leave them in the general threat intelligence backlog. Use the Threat Intelligence Feed Evaluation Guide to test a source against these requirements before buying.
Sharing identity intelligence safely
- Use TLP 2.0 labels: FIRST's Traffic Light Protocol 2.0 (authoritative since August 2022) defines TLP:CLEAR, TLP:GREEN, TLP:AMBER, TLP:AMBER+STRICT and TLP:RED. Victim-specific identity findings are usually AMBER+STRICT or RED. See the Traffic Light Protocol Guide.
- Structure the data: MISP's credential object records username, password, type (password, api-key, encryption-key), format, origin (for example leak or malware-analysis) and whether the owner or service was notified, so a recipient knows what has been done. A separate cookie object exists for session cookies. See the MISP Threat Sharing Guide.
- Exchange with STIX: STIX 2.1 is an OASIS Standard (June 2021) and is the usual format for sending campaign behaviour, indicators and relationships to partners. See the STIX and TAXII Guide.
- Notify, do not forward: when you find another organisation's secret, notify the issuer or the owner through their published channel. NIST SP 800-150 frames sharing as setting goals, scope and rules for what is published; credentials belong at the most restrictive end.
Detection and measurement
Log every identity finding with source, date first seen, matched identity, action taken and time to action. Track these metrics monthly:
- Time to revoke: median time from a credible identity finding to the credential or token being invalid. Target hours, not days.
- Match rate: percentage of findings that resolve to a known identity and owner. A low rate means the inventory, not the feed, is the problem.
- Exposure age: median age of credentials found still valid. Rising age means rotation and MFA are not keeping up.
- Repeat exposure: identities found more than once in 12 months, a sign of an infected or unmanaged device.
- Detection coverage: percentage of identity techniques from current advisories with a working detection in the SIEM or identity threat detection tooling.
Framework mapping
| Framework | Reference | How identity intelligence supports it |
|---|---|---|
| NIST CSF 2.0 | ID.RA-02, ID.RA-03, DE.AE-07 | Receiving intelligence, recording threats to identities, integrating it into alert analysis |
| NIST SP 800-53 Rev. 5 | PM-16 Threat Awareness Program, SI-5, RA-10 Threat Hunting, AC-2, IA-4, IA-5 | Threat awareness and advisories feeding account, identifier and authenticator management |
| ISO/IEC 27001:2022 | Annex A 5.7 Threat intelligence | Collecting and analysing threat information, here with identity as the subject |
| MITRE ATT&CK v20 | T1589.001, T1555.003, T1539, T1528, T1550.001, T1552.001, T1078.004, T1621 | Techniques that identity findings detect or pre-empt |
Common failure patterns
- Collecting without a join: buying stealer log coverage with no way to match findings to accounts and owners, so alerts pile up unresolved.
- Resetting the password but not the session: stealer logs carry cookies; a password reset alone leaves a stolen session working. Always revoke sessions and refresh tokens too.
- Ignoring old exposures: discarding findings older than a few months, when the Snowflake campaign used credentials exposed years earlier.
- Human-only scope: monitoring employee emails but not API keys, OAuth grants, CI tokens and service accounts, which is where much of the recent damage has come from.
- Forgotten identities: test tenants and legacy apps outside the inventory. In the Microsoft Midnight Blizzard Breach 2024, a password spray against a legacy test tenant account without MFA led to an over-privileged legacy OAuth application.
- Spreading the secret: pasting live credentials into tickets, chat and email while handling a finding, creating fresh leaks.
Lessons from incidents
- Snowflake customer campaign, 2024: about 165 organisations were notified; affected accounts had no MFA, credentials were still valid, and there were no network allow lists. Credential monitoring was one of Mandiant's core recommendations. See the Snowflake Breach 2024.
- Salesloft Drift, 2025: tokens held by a supplier were the entry point, and the actor mined stolen data for more secrets. Intelligence on a supplier compromise needs to reach the owners of every affected integration on the same day.
- Midnight Blizzard, 2024: the targets were non-production identities with no owner. Intelligence is wasted if the inventory does not include them.
Practitioner checklist
- Write identity and NHI intelligence requirements, each with an action owner and a target time to act.
- Join intelligence sources to an inventory of human and non-human identities with owners, privilege and MFA state.
- Subscribe to code host secret scanning and provider revocation alerts, and route them to a monitored queue.
- Map every SaaS integration and OAuth grant to its supplier so a vendor advisory can be actioned the same day.
- Revoke sessions and refresh tokens, not only passwords, for every stealer log finding.
- Match historical leaks against current accounts at least once, then continuously.
- Build detections for the identity techniques named in current advisories, and hunt for campaign client fingerprints.
- Share patterns under TLP 2.0 and structured formats; never forward live secrets.
- Report time to revoke, match rate and exposure age monthly.
Standards and references
- NIST: SP 800-150, Guide to Cyber Threat Information Sharing
- CISA and FBI: Scattered Spider advisory (AA23-320A)
- Google Cloud (Mandiant): UNC5537 targets Snowflake customer instances for data theft and extortion
- FIRST: Traffic Light Protocol (TLP) 2.0
- OASIS: STIX Version 2.1
- MISP Project: MISP object templates
- GitHub Docs: Secret scanning partner program
- GitHub Blog: Behind GitHub's new authentication token formats
Related NHI Mgmt Group resources: Threat Intelligence Guide · Dark Web Monitoring Guide · Infostealer Logs and Session Token Abuse Guide · Leaked Credential Response Playbook