TA419 — AI Policy AitM Credential Phishing

China-aligned cold-outreach campaign against US AI-policy experts, ending in OneDrive-themed Frameless BitB session-cookie theft
Threat
TA419 (China-aligned espionage)
Severity
ACTIVE NATION-STATE
Type
Identity / AitM Credential Phishing
Access
Cold-outreach email → AitM session-cookie theft
Version
v0.1 — 2026-10-04
Author
HuntPack
Confidence
High (1 T1 vendor primary + 4 T2 press, no contradictions)
01

Executive Summary

TA419, a China-aligned, espionage-motivated threat actor tracked by Proofpoint since at least April 2025, has run credential-phishing campaigns against US AI-policy experts at think tanks, universities, and legal-sector organizations from at least December 2025 through July 2026 (reported 2026-10-01). The attacker's objective is session-cookie / credential theft that survives MFA, not malware deployment: there is no payload anywhere in this chain.

The lure is a benign "conversation starter" email impersonating real named policy figures (a former White House OSTP leadership member, an economist/foreign-policy expert, and in one wave a senior Anthropic employee) inviting the target to join an "AI Policy Advisory Committee" or contribute to a report. Only after the victim replies does TA419 send a shortened URL that redirects through a first-stage filtering domain (Cloudflare Turnstile gate), then lands on a second-stage OneDrive-themed page built on a customized "Frameless BitB" browser-in-the-browser kit. That kit relays the victim's real Microsoft sign-in — including the MFA challenge — to genuine Microsoft infrastructure while silently capturing the resulting session cookie, giving the attacker a signed-in session without ever needing to defeat MFA directly.

Because this is a browser/authentication-layer compromise, the highest-value defensive angle is not endpoint detection — there is no process or file artifact to find — it is (1) blocking the 19 published pieces of infrastructure, (2) correlating DNS/browser-process telemetry to catch a host mid-chain, and (3) hunting Entra/M365 sign-in logs for the session-cookie-replay signature (non-interactive sign-in from a new ASN/country right after a legitimate interactive one, with no fresh MFA challenge). Phishing-resistant FIDO2/WebAuthn MFA structurally defeats this technique because the credential is bound to the legitimate origin and cannot be relayed through a look-alike domain.

Defender priority: if any of the identity hunts in §8 fire, treat it as a likely-successful compromise already in progress — revoke sessions and force re-authentication immediately. Do not wait for an endpoint signal that this campaign does not produce.
02

Source Review & Web Hunter Notes

Five sources were fetched directly and saved as plain-text snapshots (see §18 for access dates). No fetched content contained text addressed to an AI agent, instructions to fetch further URLs, claims of prior authorization, or requests to skip verification steps — all five are used as cited with no exclusions.

#SourceTierContributed
1Proofpoint — "Hallucinating Credibility" (2026-10-01)T1Full attack chain, complete 19-indicator IOC table, impersonated identities, recommendations
2The Hacker News (2026-10-04)T2Frameless BitB technical attribution, corroborates chain
3CyberScoop (2026-10-01)T2Names Lynne Parker and Heidi Crebo-Rediker as impersonated identities; NameSilo/Cloudflare infra detail
4Help Net Security (2026-10-02)T28 July 2026 campaign start date; "Keep me signed in" / OTP auto-submit kit detail; NameSilo registrar note
5IT Pro (2026-10-02)T2Corroborates MFA/session-cookie outcome; UNK_SweetSpecter comparison

All five sources agree on the core attack chain and impersonated identities; no contradictions were found. No file hashes, malware, droppers, or persistence mechanisms were published by any source — this is expected for a pure browser/authentication-layer campaign and is carried through this pack accordingly (see §15).

03

Hunt Brief & Attack Chain

Hypothesis: If TA419 is active against our user population, we should see (a) DNS/network resolution of the published infrastructure from browser processes, and/or (b) an Entra/M365 sign-in anomaly consistent with session-cookie replay, for users who received unsolicited "AI policy" themed outreach.

StepBehaviorObservable
1Benign cold-outreach email, no link/attachment, impersonating a real named policy figure or AI-company employeeMail telemetry only (sender address / impersonation domain)
2Victim replies — the trust-establishing trigger for stage 2Mail telemetry only (reply event)
3Follow-up message carries a shortened URLMail telemetry; generic shortener DNS resolution on click
4Multi-stage redirect to a first-stage TA419 domainDnsRequest / NetworkConnect from a browser process
5Cloudflare Turnstile CAPTCHA gate (filters sandboxes/crawlers)Not independently observable in endpoint telemetry
6Landing on OneDrive-themed Frameless BitB AitM page (second-stage domain)DnsRequest / NetworkConnect from a browser process — highest-fidelity endpoint signal
7AitM proxy relays credentials + MFA to real Microsoft infrastructure, captures session cookieNot observable at the endpoint; observable as an Entra/M365 sign-in anomaly
8Attacker replays stolen session cookie; signs in without fresh MFAEntra/M365 sign-in logs — non-interactive sign-in, new ASN/country, no fresh MFA claim
9Post-compromise abuse (optional): new mail rule, OAuth consent grant, mass mailbox accessEntra/M365 audit logs
04

Consolidated IOC Table

All 23 atomic indicators below are verbatim from Proofpoint's published IOC table (2026-10-01). None were invented; none exist beyond this set. Values below are de-fanged for safe display — real values are used in the CQL (§8) and IOC appendix (§10).

TypeValueGroupFirst SeenConfidenceAction
Domaindriftshare[.]coFirst-stage redirectJul 2026HighDetect
Domainquickfly[.]onlineFirst-stage redirectMay 2026HighDetect
Domaincirrushare[.]coFirst-stage redirectApr 2026HighDetect
Domaingoshshare[.]onlineFirst-stage redirectMar 2026HighDetect
Domainsynchvault[.]coFirst-stage redirectMar 2026HighDetect
Domainmsfile[.]onlineFirst-stage redirectFeb 2026HighDetect
Domainwinsync[.]cloudFirst-stage redirectFeb 2026HighDetect
Domainsharehub[.]spaceFirst-stage redirectDec 2025HighDetect
Domainglobalfileshareplatform[.]comSecond-stage AitMJul 2026HighDetect
Domainsmartsyncbox[.]comSecond-stage AitMMay 2026HighDetect
Domainmypublicshare[.]comSecond-stage AitMMar 2026HighDetect
Domaincloudsyncpulse[.]comSecond-stage AitMMar 2026HighDetect
Domainonecloudfilesync[.]comSecond-stage AitMFeb 2026HighDetect
Domainpublicsharefile[.]cloudSecond-stage AitMFeb 2026HighDetect
Domainfileswiftonline[.]cloudSecond-stage AitMDec 2025HighDetect
Domaintw-koryu[.]orgSender/impersonation (Japan-Taiwan Exchange Association)May 2026HighDetect
Domainheritiages[.]orgSender/impersonation (Heritage Foundation)Mar 2026HighDetect
Domainheritiage[.]orgSender/impersonation (Heritage Foundation)Mar 2026HighDetect
Domainshinjirou[.]infoSender/impersonation (Japanese Defense Minister)Feb 2026HighDetect
Emailleparker[at]mail[.]comAttacker-controlled senderJul 2026HighDetect / mail-gateway block
Emailhcrediker[at]mail[.]comAttacker-controlled senderJul 2026HighDetect / mail-gateway block
Emailhcrediker[at]outlook[.]comAttacker-controlled senderJul 2026HighDetect / mail-gateway block
TLS Cert SHA256b314a1499cd728ca3e54b7150661fd0c7d2279065fe3f570f0f66c395d744460Self-signed "Castro Inc" certificateFeb 2026HighHunt (where cert telemetry ingested)

No file hashes, IPs, malware, or persistence indicators are published for this campaign. None are listed here, and none should be added without a fresh, cited vendor source.

05

Affected Surface & Telemetry Matrix

SurfaceTelemetry NeededPriorityGap
Browser DNS/network (chrome.exe, msedge.exe, firefox.exe)Core Falcon: DnsRequest, NetworkConnectIP4/IP6, ProcessRollup2/SyntheticProcessRollup2HighNone — core data model
Network-layer TLS certificate metadataZeek/Corelight ssl.log, firewall TLS inspection, or CDN/proxy log ingested into NG-SIEMMediumNot in core Falcon EDR data model — tenant-dependent
Email (sender address, impersonation domain)Falcon for Email or M365 message trace / O365 Management Activity API ingested into NG-SIEMHighRequires separate email-telemetry ingestion
Entra ID / M365 sign-in & audit logsazure_signinlogs / azure_auditlogs ingested into NG-SIEM (or native Entra portal)Critical — this is where the actual compromise is detectableRequires Entra ID P1/P2 licensing for full sign-in log detail; field names are parser-prefixed per tenant
Endpoint process/file/persistenceN/AN/ANo coverage exists or is claimed — this campaign has no process, file, or persistence footprint.
06

ATT&CK Mapping

TacticTechniqueBehaviorQuery / Control
Initial AccessT1566.002 — Spearphishing LinkBenign cold-outreach email, follow-up shortened URL after replyQ4 (email hunt); M1021 hardening
Defense EvasionT1497 — Virtualization/Sandbox EvasionCloudflare Turnstile gate filters sandboxes/crawlersNot independently observable — contextual only
Credential AccessT1557 — Adversary-in-the-MiddleFrameless BitB relays live MS sign-in flowQ1, Q2 (DNS/browser-process hunts)
Credential AccessT1539 — Steal Web Session CookieAitM proxy captures session cookie post-authenticationQ5, Q6 (identity sign-in anomaly hunts)
Credential AccessT1621 — MFA Request Generation / relay (inferred)Sign-in succeeds with no fresh MFA challenge recordedQ7
Resource DevelopmentT1583.001 / T1585.002 — Acquire Infrastructure / Email AccountsDomain registration (NameSilo), disposable sender mailboxesQ1, Q2, Q4; IOC blocklist (§10)
Defense EvasionT1036.005 — Masquerading (Match Legitimate Name)Lookalike domains for Heritage Foundation, Japan-Taiwan Exchange Association, Japanese Defense MinistryQ4
(post-compromise, inferred)T1098 / T1114 — Account Manipulation / Email CollectionNew mail rule, OAuth consent grant, mass mailbox access after an anomalous sign-inQ8
07

Native Audit-Log Hunts (non-CQL)

These checks use native portal/report views rather than CQL and should be run alongside §8:

  • Entra ID → Security → Risky sign-ins / Risk detections: filter for "unfamiliar sign-in properties" and "anonymous IP address" risk types in the campaign window (Dec 2025–present).
  • Entra ID → Conditional Access → Insights and reporting: review sign-ins that satisfied CA with a token/session that was issued outside expected device/location patterns.
  • Exchange Admin Center → Mail flow → Rules: audit for newly created inbox rules or transport rules in the hours following any flagged sign-in.
  • Exchange/Defender → anti-phishing impersonation report: check for messages from the four sender-impersonation domains that bypassed existing policy.
  • Web/SWG proxy logs (if present): search for Cloudflare Turnstile challenge-page loads immediately preceding a visit to any second-stage domain — not visible in endpoint DNS/network telemetry alone.
08

CrowdStrike LogScale CQL Hunt Queries

Pick your tenant's cloud first — every "Open in Falcon" button below uses this selection.
Q1 · DNS Resolution — First-Stage TA419 Redirect Domains
CONF HIGHFP LOWCOST LOW

Looks for: a browser process resolving one of the 8 published first-stage redirect domains. FP: none expected — single-purpose phishing infrastructure with no legitimate use.

// HUNT: TA419 - T1566.002/T1204.001 - DNS Resolution of First-Stage Redirect Domain
// HYPOTHESIS: A host mid-way through the TA419 lure chain will issue a DNS request for one of the 8 published first-stage redirect domains, from a browser process.
// CONF: high  FP: low  COST: low | REQUIRES: core Falcon DNS + process telemetry
// LOOKBACK: 90d (set in the Falcon time picker; campaign infra active Dec 2025-Jul 2026)
#event_simpleName = DnsRequest
| DomainName = /^(driftshare\.co|quickfly\.online|cirrushare\.co|goshshare\.online|synchvault\.co|msfile\.online|winsync\.cloud|sharehub\.space)$/i
| rename(field=ContextProcessId_decimal, as=TargetProcessId_decimal)
| join(query={#event_simpleName = /ProcessRollup2|SyntheticProcessRollup2/}, field=TargetProcessId_decimal, key=TargetProcessId_decimal, mode=left)
| table([@timestamp, aid, ComputerName, UserName, DomainName, FileName, CommandLine])
| sort(@timestamp, order=desc)
Q2 · DNS Resolution — Second-Stage AitM Domains (highest-fidelity signal)
CONF HIGHFP LOWCOST LOW

Looks for: a browser process resolving one of the 7 published second-stage AitM (OneDrive-themed Frameless BitB) domains — a hit here means the user landed on the live credential-harvesting page. FP: none expected.

// HUNT: TA419 - T1557/T1539 - DNS Resolution of Second-Stage AitM Phishing Domain
// HYPOTHESIS: After completing the Cloudflare Turnstile gate, the browser resolves one of the 7 published second-stage AitM domains hosting the Frameless BitB kit.
// CONF: high  FP: low  COST: low | REQUIRES: core Falcon DNS + process telemetry
// LOOKBACK: 90d (set in the Falcon time picker)
#event_simpleName = DnsRequest
| DomainName = /^(globalfileshareplatform\.com|smartsyncbox\.com|mypublicshare\.com|cloudsyncpulse\.com|onecloudfilesync\.com|publicsharefile\.cloud|fileswiftonline\.cloud)$/i
| rename(field=ContextProcessId_decimal, as=TargetProcessId_decimal)
| join(query={#event_simpleName = /ProcessRollup2|SyntheticProcessRollup2/}, field=TargetProcessId_decimal, key=TargetProcessId_decimal, mode=left)
| table([@timestamp, aid, ComputerName, UserName, DomainName, FileName, CommandLine])
| sort(@timestamp, order=desc)
Q3 · TLS Certificate SHA256 — Self-Signed "Castro Inc" Reuse
CONF MEDIUMFP MEDIUMCOST LOW

Looks for: reuse of TA419's self-signed VPS certificate (subject/issuer DN C=US, ST=Kansas, L=Millsstad, O=Castro Inc, CN=CI). FP: core Falcon EDR does not natively capture TLS leaf-certificate telemetry — this card only runs where a network-layer source (Zeek/Corelight, firewall TLS inspection, CDN/proxy log) is ingested into NG-SIEM; field/repo names are placeholders until validated against your tenant.

// HUNT: TA419 - Infrastructure Reuse - Self-Signed "Castro Inc" TLS Certificate
// HYPOTHESIS: TA419's mail-sending VPS infrastructure reuses a self-signed cert on a high ephemeral port; reuse elsewhere indicates shared attacker infrastructure.
// CONF: medium  FP: medium  COST: low | REQUIRES: network-layer TLS/certificate telemetry (NOT core Falcon EDR)
// ⚠ TENANT-SPECIFIC: REPLACE_WITH_CERT_REPO and REPLACE_WITH_CERT_SHA256_FIELD are placeholders -- validate the actual repo and field names against your ingested TLS/certificate data source before relying on this query.
// TUNING: confirm the matched session is NOT a known internal TLS-inspection/proxy self-signed cert before escalating -- self-signed certs on non-standard ports have other benign uses (internal tooling, lab systems).
REPLACE_WITH_CERT_REPO
| REPLACE_WITH_CERT_SHA256_FIELD = "b314a1499cd728ca3e54b7150661fd0c7d2279065fe3f570f0f66c395d744460"
| table([@timestamp, REPLACE_WITH_SOURCE_IP_FIELD, REPLACE_WITH_DEST_IP_FIELD, REPLACE_WITH_CERT_SHA256_FIELD])
| sort(@timestamp, order=desc)
Q4 · Email — Attacker Sender Addresses & Impersonation Domains
CONF MEDIUMFP MEDIUMCOST LOW

Looks for: inbound mail from the 3 attacker-controlled sender addresses or the 4 impersonation domains. FP: exact sender-address match is high-confidence; domain match can false-positive if the tenant has legitimate correspondence with a similarly-named but unrelated organization — manually confirm the sender domain on any domain-only hit.

// HUNT: TA419 - T1566.002/T1036.005 - Attacker Sender Addresses & Impersonation Domains
// HYPOTHESIS: Inbound mail from a TA419-controlled sender address, or a domain impersonating the Heritage Foundation / Japan-Taiwan Exchange Association / Japanese Defense Ministry, lands in a mailbox.
// CONF: medium  FP: medium  COST: low | REQUIRES: Falcon for Email or M365 message trace (NOT core Falcon EDR)
// ⚠ TENANT-SPECIFIC: REPLACE_WITH_EMAIL_REPO, senderAddress, recipientAddress and subject are third-party-ingest parser field names (O365 Management Activity schema), NOT core Falcon EDR fields -- validate against your tenant's actual ingested email-telemetry repo and field names.
// TUNING: for the domain-match branch, exclude any confirmed legitimate partner mailing lists before escalating; exact sender-address hits need no exclusion.
REPLACE_WITH_EMAIL_REPO
| senderAddress = /^(leparker@mail\.com|hcrediker@mail\.com|hcrediker@outlook\.com)$/i or senderAddress = /@(tw-koryu\.org|heritiages\.org|heritiage\.org|shinjirou\.info)$/i
| table([@timestamp, senderAddress, recipientAddress, subject])
| sort(@timestamp, order=desc)
Q5 · Identity — Non-Interactive Sign-In From New Country/ASN After Interactive Sign-In
CONF MEDIUMFP MEDIUMCOST LOW

Looks for: per-user sign-in history to spot a non-interactive/token-based sign-in from a new country shortly after a legitimate interactive sign-in — the classic session-cookie-replay signature. FP: VPN egress changes, mobile carrier IP rotation, and legitimate multi-device sign-in can produce the same pattern.

// HUNT: TA419 - T1539 - Session Cookie Replay: Non-Interactive Sign-In From New Country After Interactive Sign-In
// HYPOTHESIS: After a legitimate interactive sign-in, the AitM-captured session cookie is replayed, producing a non-interactive sign-in for the same user from a different country shortly after -- with no fresh MFA challenge.
// CONF: medium  FP: medium  COST: low | REQUIRES: Entra ID sign-in logs (azure_signinlogs) -- NOT core Falcon EDR
// ⚠ TENANT-SPECIFIC: field names follow Microsoft's sign-in log schema (userPrincipalName, ipAddress, isInteractive, authenticationRequirement, location.countryOrRegion); some tenants parse these with a prefix (e.g. azuread.signinlogs.properties.*) -- validate against a sample row before relying on this query.
// TUNING: cross-check any hit against known corporate VPN egress ranges and recent business travel before escalating; collapse results per user and review the two closest-in-time entries for a country/ASN change.
// LOOKBACK: 7d (set in the Falcon time picker)
azure_signinlogs
| groupBy([userPrincipalName], function=[collect([@timestamp, ipAddress, "location.countryOrRegion", isInteractive, authenticationRequirement], limit=200)], limit=5000)
| sort(userPrincipalName, order=asc)
Q6 · Identity — Same Session/Token Identifier From Two Distinct Source IPs
CONF MEDIUMFP MEDIUMCOST LOW

Looks for: a single sign-in token identifier observed from more than one source IP — direct evidence of a stolen session being used from attacker infrastructure alongside the legitimate user. FP: corporate load-balanced egress or carrier-grade NAT can occasionally present as multiple IPs for one session.

// HUNT: TA419 - T1539 - Same Session Token From Two Distinct Source IPs
// HYPOTHESIS: A session/token identifier captured by the AitM proxy is replayed from attacker infrastructure while the legitimate user continues to use the same token -- both source IPs appear against one token in the same window.
// CONF: medium  FP: medium  COST: low | REQUIRES: Entra ID sign-in logs (azure_signinlogs) with a token/session identifier field -- NOT core Falcon EDR
// ⚠ TENANT-SPECIFIC: uniqueTokenIdentifier is Microsoft's documented sign-in log field name; confirm it is populated and not parser-renamed in your tenant.
// TUNING: exclude known NAT/load-balancer egress pools (CGNAT, corporate proxy ranges) before escalating; a DistinctIPs count of exactly 2 with one IP geolocating outside the user's normal country is the highest-confidence case.
// LOOKBACK: 7d (set in the Falcon time picker)
azure_signinlogs
| groupBy([uniqueTokenIdentifier], function=[count(ipAddress, distinct=true, as=DistinctIPs), collect([@timestamp, userPrincipalName, ipAddress, "location.countryOrRegion"], limit=50)], limit=5000)
| DistinctIPs > 1
| sort(DistinctIPs, order=desc)
Q7 · Identity — Sign-In Success Without Fresh MFA Challenge
CONF LOWFP HIGHCOST LOW

Looks for: sign-ins satisfied by a single factor (no fresh MFA claim) — on its own this is broad (legitimate CA-trusted-location or remembered-device sign-ins look identical), so it is meant to be reviewed as a corroborating signal alongside Q5/Q6, not escalated alone. FP: high — this is the expected shape of most "remember this device" sign-ins; do not alert on this query in isolation.

// HUNT: TA419 - T1621 - Sign-In Success Without Fresh MFA Challenge
// HYPOTHESIS: A session-cookie-replay sign-in succeeds without the user re-satisfying MFA, because the captured cookie already carries a satisfied authentication claim.
// CONF: low  FP: high  COST: low | REQUIRES: Entra ID sign-in logs (azure_signinlogs) -- NOT core Falcon EDR
// ⚠ TENANT-SPECIFIC: authenticationRequirement values ("singleFactorAuthentication"/"multiFactorAuthentication") follow Microsoft's documented schema -- validate against a sample row in your tenant.
// TUNING: this query alone is NOT actionable -- CA-trusted-location and "remember MFA for N days" sign-ins look identical. Only escalate results that also appear in Q5 or Q6 (new country/ASN, or a token seen from two IPs) in the same time window.
// LOOKBACK: 7d (set in the Falcon time picker)
azure_signinlogs
| authenticationRequirement = "singleFactorAuthentication"
| conditionalAccessStatus = "success"
| groupBy([userPrincipalName], function=[count(as=Hits), collect([@timestamp, ipAddress, "location.countryOrRegion", appDisplayName], limit=50)], limit=5000)
| sort(Hits, order=desc)
Q8 · Identity — Post-Sign-In OAuth Consent Grant / New Mail Rule
CONF MEDIUMFP MEDIUMCOST LOW

Looks for: post-compromise follow-on abuse — an OAuth application consent grant or a new mailbox/transport rule created shortly after a sign-in — common next steps once a session is hijacked. FP: legitimate app onboarding and admin-driven mail-rule changes will also match; corroborate against a flagged user from Q5/Q6/Q7 before escalating.

// HUNT: TA419 - T1098/T1114 (inferred) - Post-Sign-In OAuth Consent Grant or New Mail Rule
// HYPOTHESIS: Following a successful session-cookie-replay sign-in, the attacker grants an OAuth app consent or creates a mail-forwarding/inbox rule to persist access to mailbox content.
// CONF: medium  FP: medium  COST: low | REQUIRES: Entra ID / Exchange audit logs (azure_auditlogs) -- NOT core Falcon EDR
// ⚠ TENANT-SPECIFIC: operationName is an Entra/Exchange audit parser field, not a core Falcon EDR field; it and its activity-name values ("Consent to application", "Add delegated permission grant", "New-InboxRule") follow Microsoft's documented audit-log schema -- validate against a sample row in your tenant.
// TUNING: exclude known IT-managed app deployment windows and admin-initiated mail-rule changes (check the Actor field against your IT admin group) before escalating.
// LOOKBACK: 7d (set in the Falcon time picker)
azure_auditlogs
| operationName = /^(Consent to application|Add app role assignment|Add delegated permission grant|New-InboxRule|Set-Mailbox)$/i
| table([@timestamp, userPrincipalName, operationName, appDisplayName])
| sort(@timestamp, order=desc)
Q9 · Behavioral Pivot — Shortener Resolution Then Non-Microsoft OneDrive-Themed Landing (Same Host)
CONF HIGH*FP MEDIUMCOST LOW

Looks for: the same host resolving a generic public URL shortener and, separately, a published second-stage AitM domain — approximating the "reply → shortener → fake OneDrive" behavioral chain at the DNS layer. The shortener list is a generic set of widely-abused public services, not TA419-specific indicators. FP: medium — the join is scoped by host (aid) only, not a time window, so confirm the two timestamps are close together before escalating; *confidence is high only when both conditions genuinely correlate in time for the same user.

// HUNT: TA419 - Behavioral Pivot - Shortener Resolution + Non-Microsoft OneDrive-Themed Landing
// HYPOTHESIS: A user replies to an unsolicited external email, resolves a URL shortener, then lands on one of the 7 published TA419 second-stage AitM domains -- the attack chain's distinctive DNS fingerprint.
// CONF: high when both events correlate in time for one host; FP: medium  COST: low | REQUIRES: core Falcon DNS telemetry
// TUNING: this join is scoped by aid only (no in-query time window) -- manually confirm @timestamp and AitMTimestamp are within the same browsing session (minutes, not days) before treating a hit as a true positive. Shorteners have legitimate business use on their own.
// LOOKBACK: 30d (set in the Falcon time picker)
#event_simpleName = DnsRequest
| DomainName = /^(bit\.ly|tinyurl\.com|rebrand\.ly|cutt\.ly|is\.gd|t\.co|shorturl\.at)$/i
| select([@timestamp, aid, ComputerName, UserName, DomainName])
| join(query={
    #event_simpleName = DnsRequest
    | DomainName = /^(globalfileshareplatform\.com|smartsyncbox\.com|mypublicshare\.com|cloudsyncpulse\.com|onecloudfilesync\.com|publicsharefile\.cloud|fileswiftonline\.cloud)$/i
    | rename(field=DomainName, as=AitMDomain)
    | rename(field=@timestamp, as=AitMTimestamp)
    | select([aid, AitMDomain, AitMTimestamp])
  }, field=aid, key=aid, mode=inner)
| table([@timestamp, aid, ComputerName, UserName, DomainName, AitMDomain, AitMTimestamp])
| sort(@timestamp, order=desc)
09

Custom IOA Recommendations

Behavioral/prevention coverage for this campaign is limited at the endpoint because the technique is DNS/browser-layer, not process-layer. One Custom IOA is recommended; the rest of the defeat lives in Conditional Access (§12), not endpoint prevention rules.

Custom IOA Rule: TA419 Infrastructure DNS Resolution

  • Rule Group: Phishing / AitM Defense
  • Rule Type: DNS Request
  • Action: Detect (always start here; promote to Block only after a confirmed-zero FP rate over 14+ days per the standard promotion path)
  • Severity: High
  • MITRE Technique: T1557 — Adversary-in-the-Middle
  • Detection Logic — Domain Name (PCRE): ^(driftshare\.co|quickfly\.online|cirrushare\.co|goshshare\.online|synchvault\.co|msfile\.online|winsync\.cloud|sharehub\.space|globalfileshareplatform\.com|smartsyncbox\.com|mypublicshare\.com|cloudsyncpulse\.com|onecloudfilesync\.com|publicsharefile\.cloud|fileswiftonline\.cloud)$
  • FP Tuning Notes: none expected — newly-registered, single-purpose phishing infrastructure with no legitimate use. If the campaign rotates infrastructure again, this list must be refreshed from the next vendor update (see §17 re-hunt trigger).
  • Recommended Validation: from an isolated test endpoint, perform an nslookup against one listed domain (does not require loading the page) and confirm the rule fires in Detect mode before any promotion consideration.
10

Machine-Readable IOC Appendix

Grouped quick-copy blocks below use real (fanged) values for direct action. The Falcon IOC Management CSV covers the 19 domains only — Falcon IOC Management does not natively import email addresses or certificate fingerprints as indicator types; those route to mail-gateway and TLS-inspection tooling respectively.

Falcon IOC Management CSV — All 19 Domainsbulk import, detect-only
type,value,action,severity,expiration,description,tags
domain,driftshare.co,detect,high,2027-01-02,TA419 first-stage redirect domain,campaign:TA419
domain,quickfly.online,detect,high,2027-01-02,TA419 first-stage redirect domain,campaign:TA419
domain,cirrushare.co,detect,high,2027-01-02,TA419 first-stage redirect domain,campaign:TA419
domain,goshshare.online,detect,high,2027-01-02,TA419 first-stage redirect domain,campaign:TA419
domain,synchvault.co,detect,high,2027-01-02,TA419 first-stage redirect domain,campaign:TA419
domain,msfile.online,detect,high,2027-01-02,TA419 first-stage redirect domain,campaign:TA419
domain,winsync.cloud,detect,high,2027-01-02,TA419 first-stage redirect domain,campaign:TA419
domain,sharehub.space,detect,high,2027-01-02,TA419 first-stage redirect domain,campaign:TA419
domain,globalfileshareplatform.com,detect,high,2027-01-02,TA419 second-stage AitM phishing domain,campaign:TA419
domain,smartsyncbox.com,detect,high,2027-01-02,TA419 second-stage AitM phishing domain,campaign:TA419
domain,mypublicshare.com,detect,high,2027-01-02,TA419 second-stage AitM phishing domain,campaign:TA419
domain,cloudsyncpulse.com,detect,high,2027-01-02,TA419 second-stage AitM phishing domain,campaign:TA419
domain,onecloudfilesync.com,detect,high,2027-01-02,TA419 second-stage AitM phishing domain,campaign:TA419
domain,publicsharefile.cloud,detect,high,2027-01-02,TA419 second-stage AitM phishing domain,campaign:TA419
domain,fileswiftonline.cloud,detect,high,2027-01-02,TA419 second-stage AitM phishing domain,campaign:TA419
domain,tw-koryu.org,detect,high,2027-01-02,TA419 sender/impersonation domain (Japan-Taiwan Exchange Association),campaign:TA419
domain,heritiages.org,detect,high,2027-01-02,TA419 sender/impersonation domain (Heritage Foundation),campaign:TA419
domain,heritiage.org,detect,high,2027-01-02,TA419 sender/impersonation domain (Heritage Foundation),campaign:TA419
domain,shinjirou.info,detect,high,2027-01-02,TA419 sender/impersonation domain (Japanese Defense Minister),campaign:TA419
First-Stage Redirect Domains8 values
driftshare.co
quickfly.online
cirrushare.co
goshshare.online
synchvault.co
msfile.online
winsync.cloud
sharehub.space
Second-Stage AitM Domains7 values
globalfileshareplatform.com
smartsyncbox.com
mypublicshare.com
cloudsyncpulse.com
onecloudfilesync.com
publicsharefile.cloud
fileswiftonline.cloud
Sender / Impersonation Domains4 values
tw-koryu.org
heritiages.org
heritiage.org
shinjirou.info
Attacker Email Addresses3 values
leparker@mail.com
hcrediker@mail.com
hcrediker@outlook.com
TLS Certificate SHA2561 value — "Castro Inc" self-signed
b314a1499cd728ca3e54b7150661fd0c7d2279065fe3f570f0f66c395d744460
11

Detection Validation Gates

GateCheckStatus Before Deploy
Telemetry readyDnsRequest/NetworkConnect/ProcessRollup2 flowing for the browser fleet (Q1/Q2); azure_signinlogs/azure_auditlogs ingested or Entra portal access confirmed (Q5-Q8)Confirm before relying on any identity-hunt result
Benign baselineRun Q1/Q2/Q9 over a known-clean 7-day window; expect zero hits (newly-registered single-purpose domains)Any hit in a clean baseline window warrants immediate investigation, not dismissal
Positive testQ1-Q4, Q9: nslookup/DNS-only test against a listed domain confirms the query fires (do not browse to the live phishing page)Safe to test in an isolated sensor group
PromotionCustom IOA (§9) promoted from Detect to Block only after 14+ days with a measured 0% FP rate on a canary sensor groupNot yet promoted — ships in Detect mode
12

Hardening & Prevention

TTP: TA419 AI-Policy AitM Credential Phishing (Frameless BitB / session-cookie theft)   MITRE Technique: T1557, T1539, T1566.002, T1621   Primary Mitigations: M1032, M1056, M1017, M1021, M1018, M1047

This compromise lands entirely on the identity/authentication layer — there is no endpoint payload to contain. Hardening centers on making a captured credential/session cookie worthless (phishing-resistant auth, token binding) and on narrowing the lure's delivery surface.

Immediate — deploy this week
  • Block the full published TA419 indicator set (19 domains at the web proxy and DNS/firewall; the 3 attacker sender addresses at the mail gateway; the TLS cert fingerprint wherever cert telemetry is inspected) — 23 indicators in total. Framework: M1021; Proofpoint primary source (2026-10-01).
  • Revoke refresh tokens / force session sign-out for any user who resolved a listed domain, invalidating any session cookie already captured. Framework: M1018; Microsoft Entra "Revoke user sessions."
  • Enable Continuous Access Evaluation (CAE) tenant-wide to shorten the usable window of a stolen token. Framework: M1032; Microsoft Entra Conditional Access / CAE documentation.
  • Enable impersonation-protection / external-sender banners for the four lookalike partner domains and the organizations they imitate (Heritage Foundation, Japan-Taiwan Exchange Association, Japanese Defense Ministry). Framework: M1021; Exchange Online Protection anti-phishing policy.
Short-term — 1 to 4 weeks
  • Roll out phishing-resistant FIDO2/WebAuthn passkeys to the highest-risk population first (AI-policy, government-relations, executive staff) — the single control that structurally defeats Frameless BitB AitM. Framework: M1032; NIST SP 800-63B (AAL3); Microsoft Entra passkey/FIDO2 deployment guidance.
  • Deploy Conditional Access token protection (sign-in session binding) so an issued token can only be redeemed from the device it was issued to. Framework: M1032; Microsoft Entra Conditional Access "Token protection."
  • Require compliant-device / managed-browser Conditional Access for Microsoft 365 access, closing the gap where a stolen cookie is replayed from an unmanaged attacker browser. Framework: M1018.
  • Add shortened-URL expansion and newly-registered-domain scoring at the mail/web gateway — stage 2 always arrives as a shortened URL, and TA419's domains are registered through NameSilo and fronted by Cloudflare CDN to hide the backend host, so registrar/CDN-pattern and domain-reputation scoring both apply. (Proofpoint publishes first-seen dates per domain, not registration dates, so age-based scoring should be tuned on your own WHOIS feed rather than an assumed domain age.) Framework: M1021; CIS web-content-filtering control.
Long-term / structural — 1 to 3 months
  • User training on the "benign conversation-starter" pretext specifically — this campaign's defining trait defeats "don't click suspicious links" training because stage 1 has no link. Train on verifying unsolicited outreach via an independent channel before replying. Framework: M1017.
  • Expand token-protection and passkey enforcement tenant-wide, retiring SMS/voice/push MFA fallback where feasible since both remain relayable by an AitM proxy. Framework: M1032; NIST SP 800-63B.
  • Operationalize recurring re-ingestion of TA419 IOC updates and re-run this pack's identity hunts on a recurring cadence given the campaign's monthly infrastructure rotation. Framework: M1047.

What to verify after deployment

  • Confirm the 19 domains return a block/deny verdict at the web proxy and the 3 sender addresses are rejected at the mail gateway (test against the blocklist directly, not by browsing to the live phishing domains).
  • Confirm CAE is active: Entra ID → Security → Conditional Access → Insights and reporting should show "Continuous access evaluation" enabled.
  • Confirm token-protection policy state via Microsoft Graph PowerShell (Get-MgIdentityConditionalAccessPolicy) shows the session control attached.
  • Re-run Q5-Q8 (§8) 7 and 30 days after rollout to confirm the anomaly pattern has not recurred for enrolled users.
  • Confirm passkey/FIDO2 enrollment rate via Entra ID → Authentication methods → Registration report.
13

Deployable Playbooks

MITRE: T1557/T1539 → M1032, M1018   Estimated deploy time: 2-4 hours for immediate tier   Prerequisites: Entra ID P1/P2, Microsoft Graph PowerShell SDK

Step 1 — Revoke sessions for a suspected-compromised user

What it does: invalidates all refresh tokens and active sessions for one user. Why it works: an AitM-captured session cookie is worthless once the underlying refresh token is revoked (M1018).

# Requires: Connect-MgGraph -Scopes "User.RevokeSessions.All"
Revoke-MgUserSignInSession -UserId "REPLACE_WITH_USER_UPN_OR_ID"

# Verify:
Get-MgAuditLogSignIn -Filter "userPrincipalName eq 'REPLACE_WITH_USER_UPN'" -Top 5

Step 2 — Enable Conditional Access token protection

What it does: binds an issued token to the device/session it was granted to. Why it works: an AitM proxy capturing the token cannot replay it from a different device (M1032).

# Portal path:
# Entra admin center > Protection > Conditional Access > Policies > New policy
#   Target resources: Office 365 (or all cloud apps)
#   Session > Token protection > Require token protection for sign-in sessions

# Verify (Graph PowerShell):
Get-MgIdentityConditionalAccessPolicy | Where-Object { $_.SessionControls.PersistentBrowser -ne $null -or $_.SessionControls.AdditionalProperties.tokenProtection -ne $null }

Step 3 — Block the published IOC set at the mail/web gateway

What it does: prevents delivery/resolution of TA419's known infrastructure. Why it works: removes the lure and the landing page from reachability (M1021).

# Generic pattern -- adapt to your gateway's CLI/API (example shown for a generic blocklist import):
REPLACE_WITH_GATEWAY_CLI import-blocklist --type domain --file ta419_domains.txt --action block
REPLACE_WITH_GATEWAY_CLI import-blocklist --type sender --file ta419_senders.txt --action block

# Verify: attempt a controlled, non-browsing DNS lookup for one listed domain from a managed network path
# and confirm the gateway logs a block/deny entry.

Step 4 — Enroll high-risk users in FIDO2/passkey authentication

What it does: issues origin-bound, phishing-resistant credentials. Why it works: the credential cannot be relayed through a look-alike domain, structurally defeating Frameless BitB (M1032).

# Portal path:
# Entra admin center > Protection > Authentication methods > Policies > FIDO2 security key > Enable
# Target the high-risk group (AI-policy / government-relations / executive staff) first.

# Verify:
Get-MgReportAuthenticationMethodUserRegistrationDetail -Filter "userPrincipalName eq 'REPLACE_WITH_USER_UPN'"
14

Containment Runbook

PhaseActionsOwnerEvidence
1. Immediate (0-1 hr)Revoke sessions, force password reset for affected user(s)IR / Identity teamEntra sign-in log export, revocation timestamp
2. Short-term (1-24 hr)Block all 23 published indicators (19 domains, 3 sender addresses, 1 TLS cert fingerprint) at the relevant gateway; notify user via a verified separate channel (not the original email thread); review mailbox for new rules/OAuth grantsSOC / Messaging teamGateway block confirmation, mailbox audit export
3. Recovery (24-72 hr)Re-enroll user in FIDO2/passkey; confirm CA token-protection policy applied to the accountIAM teamAuthentication-methods registration report
4. Post-incident (1-2 wk)Re-run this pack's identity hunts (Q5-Q8); update detections; brief affected user population on the pretext patternThreat Hunting teamRe-hunt results, training attendance record
15

Detection Coverage Map

MITRE TechniqueTechnique NameCQL QueryCustom IOACoverage
T1566.002Spearphishing LinkQ4—Partial
T1204.001 / T1557AitM via malicious linkQ1, Q2IOA-1Good
T1497Sandbox evasion (Turnstile)——GAP
T1539Steal Web Session CookieQ5, Q6—Partial
T1621MFA bypass via relayQ7—Partial
T1036.005Masquerading (lookalike sender domains)Q4—Partial
T1098/T1114 (post-compromise)Account manipulation / email collectionQ8—Partial
Infrastructure reuse (TLS cert)—Q3—Partial

Coverage summary: 7 of 8 identified techniques have at least partial detection coverage; 1 gap (Turnstile sandbox evasion, not independently observable in any available telemetry).

There is no process-, file-, or persistence-level telemetry for this campaign, and none is claimed. Proofpoint published no malware, droppers, executables, or payload hashes — this is purely a browser/authentication-layer compromise. Endpoint-side coverage in this pack is limited to DNS/network resolution of the 19 listed infrastructure indicators correlated to browser process context (chrome.exe, msedge.exe, firefox.exe). Full detection of the AitM/session-theft outcome requires identity (Entra/M365 sign-in log) and email telemetry that sit outside the core Falcon endpoint data model — these are explicitly marked ⚠ tenant-specific throughout §8 and must be validated against each tenant's actual ingested schema before use.

16

Hunt Summary Ticket

TITLE:     TA419 AI-Policy AitM Credential Phishing (Frameless BitB)
SEVERITY:  HIGH (active nation-state campaign; identity-layer MFA-bypass outcome)
SCOPE:     US think tanks, universities, legal-sector orgs; AI-policy experts specifically targeted
HYPOTHESIS: Browser DNS/network resolution of 15 published TA419 domains, or an Entra/M365
           sign-in anomaly consistent with session-cookie replay, indicates an active or
           already-successful compromise.
QUERIES:   Q1-Q9 (section 8) -- Q2 (second-stage AitM domain resolution) and Q5/Q6
           (identity session-replay) are highest priority.
DO FIRST:  Block all 19 published indicators at mail/web/DNS gateway; run Q1/Q2 against the
           last 90 days; run Q5-Q8 against the last 7 days for any AI-policy-adjacent user group.
FINDINGS:  Queries must be run manually in Falcon/Entra; this ticket does not assert results.
GAPS:      No process/file/persistence telemetry exists for this campaign (none published).
           Cloudflare Turnstile gate is not independently observable in any available telemetry.
ACTIONS:   See section 14 (Containment Runbook) for phased response if any query fires.
OWNER:     REPLACE_WITH_ANALYST_NAME
VERSION:   v0.1 -- 2026-10-04
17

Changelog

v0.12026-10-04Initial pack build from Proofpoint primary source (2026-10-01) plus 4 corroborating press sources. 9 CQL query cards, 1 Custom IOA, tiered hardening, containment runbook. Re-hunt trigger: new Proofpoint/vendor IOC update, or confirmed infrastructure rotation beyond the 19 published indicators.
18

References

TierSourceUsed ForAccess Date
T1Proofpoint — "Hallucinating Credibility: China-Aligned TA419 Impersonates its Way into US AI Policy Circles"Full attack chain, complete IOC table, recommendations2026-10-04
T2The Hacker NewsFrameless BitB attribution detail2026-10-04
T2CyberScoopImpersonated-identity names, infrastructure detail2026-10-04
T2Help Net SecurityCampaign start date, kit automation detail2026-10-04
T2IT ProCorroboration, comparison to UNK_SweetSpecter2026-10-04
FrameworkMITRE ATT&CK T1557 — Adversary-in-the-MiddleTechnique mapping2026-10-04
FrameworkMITRE ATT&CK T1539 — Steal Web Session CookieTechnique mapping2026-10-04
StandardNIST SP 800-63BPhishing-resistant authenticator guidance (AAL3)2026-10-04

Source snapshots (plain text, saved verbatim at fetch time) are stored alongside this pack in TA419-AIPolicy-AiTM-Hunt-sources/ for IOC-provenance verification.