Most brute force playbooks still optimize for one noisy IP hammering one login form. That pattern is easy to rate-limit and easy to celebrate in a dashboard. The campaigns that actually land accounts look quieter: two or three guesses per exit, dozens of cloud and residential sources, the same username list reused across SSH, VPN, and the corporate IdP within the same hour.
If your detection story is "N failures from one address equals block," those campaigns read as healthy noise. Your lockout counters stay calm. Your ASN caps look unused. Helpdesk tickets stay flat until an account that already had weak MFA or a reused password starts behaving like an insider.
Why single-IP thresholds keep missing the real campaign
Classic controls were built for dial-up era guessing. Fail2ban-style bans, WAF bot rules, and IdP lockouts all assume concentration. Distributed guessing breaks that assumption on purpose.
Attackers pull exits from the same pools that show up in weekly threat reports: cloud tenants used as disposable scanners, consumer VPN egress, and compromised CPE or set-top paths that rotate through friendly geographies. Operation ShadowCloud-style tradecraft already showed how cloud service paths blend into legitimate ASN space. Brute force operators borrow the same cover. One attempt per address looks like a fat-fingered user. Twenty addresses aiming at the same privileged mailbox looks like weather.
Teams also split telemetry by product. SSH failures live in host logs. RDP failures sit in Windows Security. IdP failures live with the identity vendor. VPN failures stay with the remote access appliance. Each silo sees a trickle. Nobody owns the join that turns those trickles into a credential hunt.
Threat intelligence spend does not fix that join by itself. Recent surveys asking what intel is worth keep landing on the same answer from practitioners: value shows up when a feed changes a same-shift decision. A blocklist of yesterday's noisy /24 helps after the spray already rotated. Velocity and credential reuse across your own auth surfaces close the gap faster.
What "low and slow" looks like in production logs
Treat these shapes as investigation triggers, not as proof of compromise on their own.
- Per-source thinness, per-identity thickness: each source IP fails two or fewer times against a target, while one username or UPNs accumulates failures from many sources in under sixty minutes.
- Protocol hopping: the same identity string appears in SSH auth.log, then Microsoft Entra or Okta sign-in logs, then a VPN concentrator within a short window.
- Clockwork spacing: attempts arrive every few minutes with near-uniform intervals, especially overnight when legitimate staff retry rates drop.
- Shared tooling fingerprints: identical user-agent strings, TLS JA3/JA4 clusters, SSH client versions, or RDP cookie patterns across unrelated ASNs.
- Password-adjacent sequencing: failures that walk seasons, keyboard walks, or breach-dump variants against a short VIP list rather than spraying the full directory at once.
Contrast that with a helpdesk-driven lockout storm. Those usually cluster around a password reset wave, a broken SSO connector, or a single office subnet after a cert expiry. Brute force campaigns scatter by ASN and concentrate by identity.
A same-hour playbook that works without boiling the ocean
1. Build one auth failure fact table
Normalize every failed authentication into a narrow schema your SOC can join hourly:
- timestamp (UTC)
- identity (normalized UPN or local account)
- source IP, ASN, and prefix
- protocol or app (ssh, rdp, vpn, idp, mail, admin portal)
- result code and MFA step if present
- client fingerprint when available
Ship that join into the SIEM or a warehouse table with a sixty-minute sliding window. Analysts should be able to ask "which identities saw failures from six or more distinct /24s in the last hour?" without opening four consoles.
2. Score campaigns, then score addresses
Invert the usual priority. First flag the identity or role under pressure. Then rank contributing sources.
Useful campaign scores for an identity window:
- distinct source prefixes
- distinct ASNs
- distinct protocols
- failure-to-success ratio
- overlap with known spray wordlists or recent breach dump usernames
Only after the identity crosses a threshold should automatic controls start quarantine actions on the contributing prefixes. That order keeps a busy executive who mistypes a password three times from triggering a site-wide ban, while still catching the twenty-exit guess against the same finance shared mailbox.
3. Cap by identity and by role, with different ceilings
Apply tighter velocity limits to privileged roles than to standard staff. Break-glass, domain admins, cloud owners, finance shared inboxes, and remote support accounts deserve lower allowed failure diversity even when MFA is on.
Implementation pattern many IdPs and reverse proxies already support:
- soft challenge after a small per-identity failure burst
- temporary step-up MFA or number matching for privileged roles
- prefix quarantine at the edge for sources that contribute to a multi-ASN identity campaign
- account soft-hold that requires helpdesk attestation rather than a hard lock that creates a self-DoS for the attacker
Hard lockouts remain useful for local accounts that lack MFA. For federated users, soft holds plus source quarantine usually preserve availability better.
4. Put decoys where guessing is cheap
Plant a small set of honey accounts with realistic names and no production rights. Alert on any authentication attempt against them across SSH, RDP, and IdP. A single hit from a new ASN against a honey UPN is a cleaner signal than the fiftieth failure against a real user who just returned from vacation.
Pair those accounts with a canary admin path or SSH port that only your monitoring knows. Mass scouts and ransomware staging often knock decoy services in the same week as credential guessing. Keep those hunts separate so PHP mass scouts do not drown the credential campaign ticket.
5. Rehearse the containment path before dump week
Credential stuffing and classic brute force both spike when large breach dumps land. Schedule a short tabletop around those release windows:
- Who owns the identity-campaign alert?
- How fast can edge quarantine consume a prefix list from the SIEM?
- What is the failback if a scrubber or WAF rate rule adds auth latency?
- Which partner SSO apps still accept password fallback and therefore need temporary deny?
Coinbase-scale social engineering headlines remind leadership that users will click deepfake prompts. Your job on the auth side is making sure the parallel technical path, password guessing against exposed surfaces, has nowhere quiet to land.
Detection rules you can implement this week
Keep rules boring and measurable.
- Identity multi-prefix: alert when one identity sees failures from ≥6 distinct /24s in 60 minutes across any two protocols.
- Privileged thin-source: alert when a privileged role sees ≥3 distinct ASNs with ≤2 failures each in 30 minutes.
- Protocol hop: alert when the same identity fails SSH or RDP and then IdP or VPN within 45 minutes from different prefixes.
- Honey account any-touch: page on first failure or success attempt against honey identities.
- Success after diversity: critical alert when a multi-prefix failure campaign against an identity is followed by a success from a new ASN within two hours.
Tune thresholds with a two-week baseline of your own helpdesk and travel patterns. Remote support ASNs and partner SOCs need allow-listed ceilings so legitimate jump hosts do not look like spray infrastructure.
Prevention that still leaves remote work intact
Detection without prevention just produces prettier tickets. Prefer controls that raise attacker cost without trapping travelers.
- Disable password auth on internet-facing SSH where keys or short-lived certificates are viable.
- Move admin portals behind SSO with phishing-resistant MFA; keep password endpoints off the public internet.
- Enforce unique, long local passwords for service accounts and remove interactive logon where possible.
- Rate-limit authentication endpoints by identity and by /24, not only by single IP.
- Reject auth traffic from ASNs with elevated abuse velocity against your honey accounts, even when commercial reputation scores still disagree.
- For APIs, require attested clients or mTLS on token endpoints so credential stuffing scripts cannot look like first-party apps.
Phishing pages that harvest MX records and brand lookalikes still feed the password lists that later show up in guessing campaigns. Treat inbox phishing controls and auth velocity controls as one lifecycle: steal, spray, then reuse.
A concrete Monday morning sequence
- Pull the top twenty identities by distinct failure prefixes over the last 24 hours.
- Cross-check those identities against privileged roles and shared mailboxes.
- For each elevated hit, export contributing ASNs and compare to honey-account touch lists.
- Quarantine the top contributing prefixes at the edge for 24 hours with an exception path for known support ASNs.
- Force a password and session review on any account that saw success after multi-prefix failure.
- File one hunting packet: identity, timeline, protocols, prefixes, and whether the password appeared in a recent dump.
That sequence beats waiting for a country deny or a permanent blocklist entry to "catch up" after the exits already rotated.
Takeaways worth pinning on the SOC wall
- Brute force that matters is often thin per IP and thick per identity.
- Join auth failures across SSH, VPN, RDP, and IdP in the same hour or you will under-detect by design.
- Score the campaign first, then the addresses that feed it.
- Privileged roles need lower diversity ceilings than standard users.
- Honey accounts turn quiet guessing into high-confidence tickets.
- Edge quarantines keyed to identity campaigns age better than single-IP bans during cloud and residential exit churn.
If your graphs show "brute force blocked" based only on loud single sources, assume the quiet campaign is already in the logs. The work is teaching your stack to notice two knocks from twenty doors aimed at the same key.