Skip to navigation

Threat reportPhishingTL-2026-1963

AI Voice-Cloning Vishing Wave Hits Point72, Citadel, Two Sigma, Millennium Management — Tied to UNC6671 (BlackFile/Redact) Extortion Group

highACTIVE

On August 5, 2026, a coordinated wave of AI voice-cloning vishing attacks targeted Point72 Asset Management, Citadel, Two Sigma Investments, and Millennium Management, with attackers impersonating IT help-desk staff and colleagues to trick employees/help-desk agents into resetting credentials and granting system access.

Key facts

IdentifiersTL-2026-1963
Also known asWall Street Hedge Fund Vishing Wave, Point72 AI Vishing Incident, Hedge Fund AI Voice-Cloning Campaign
SeverityHIGH
Affected productsN/A (social-engineering/identity-process attack, not a software vulnerability) Okta / Microsoft Entra ID SSO, Microsoft 365, Google Workspace, and connected SaaS/OAuth-integrated applications (e.g. Salesforce ecosystem)
AttributionUNC6671 (medium confidence)
MalwareBlackFile, Citadel, Falcon, Helix, Pink, Redact
MITRE ATT&CK techniquesT1020 Automated Exfiltration, T1071 Application Layer Protocol, T1087 Account Discovery, T1098 Account Manipulation, T1110 Brute Force, T1526 Cloud Service Discovery, T1531 Account Access Removal, T1537 Transfer Data to Cloud Account, T1550 Use Alternate Authentication Material, T1556 Modify Authentication Process, T1557 Adversary-in-the-Middle, T1566 Phishing and 2 more in the attack flow
MotivationFINANCIAL
Target sectorsfinancial services, hedge funds, asset management, private equity, legal services, financial rating agencies
Target regionsNorth America, united states of america
Detection rules9 (1 shown free)
Sources15 cited sources

Detection coverage

As of 2026-08-09, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-1963 across Splunk SPL, Microsoft KQL and Sigma, covering 25 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: Automated Non-Human SaaS Bulk Data Exfiltration to External Egress Infrastructure (UNC6671)

Detects the exfiltration stage of the UNC6671 kill chain: script-driven bulk downloads/exports from Microsoft 365/SharePoint/Okta-connected SaaS using non-human User-Agent strings (python-requests, WindowsPowerShell, anomalous Okta mobile client builds) at high volume, and/or traffic to UNC6671's known SaaS-egress IPs — the automated transfer of stolen data to an attacker-controlled cloud account ahead of brand-specific data-leak-site extortion.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1087, T1526, T1537, T1020, T1071
  • Data sources: o365_management_activity, sharepoint_audit_logs, okta_system_log, cloud_api_gateway_logs
  • Severity: critical
  • Confidence: high

Splunk SPL

index=o365 sourcetype="o365:management:activity"
Operation IN ("FileDownloaded", "FileAccessed", "FileSyncDownloadedFull", "FilePreviewed", "SearchExported")
(UserAgent="python-requests/2.28.1" OR UserAgent="*python-requests*" OR UserAgent="WindowsPowerShell/5.1" OR UserAgent="*PowerShell*" OR ClientIP IN ("23.234.75.84","195.140.213.114","195.140.213.115") OR UserAgent="0811A9866E.com.okta.android.auth/8.18.0")
| bin _time span=1h
| stats count dc(SourceFileName) as distinct_files values(Operation) as operations values(UserAgent) as user_agents by UserId, ClientIP, _time
| where count>=50 OR distinct_files>=20
| sort -count

Microsoft KQL

OfficeActivity
| where TimeGenerated > ago(24h)
| where Operation in ("FileDownloaded","FileAccessed","FileSyncDownloadedFull","FilePreviewed","SearchExported")
| where UserAgent has_any ("python-requests","PowerShell") or ClientIP in ("23.234.75.84","195.140.213.114","195.140.213.115") or UserAgent == "0811A9866E.com.okta.android.auth/8.18.0"
| summarize EventCount=count(), DistinctFiles=dcount(SourceFileName), Operations=make_set(Operation), UserAgents=make_set(UserAgent) by UserId, ClientIP, bin(TimeGenerated, 1h)
| where EventCount >= 50 or DistinctFiles >= 20
| order by EventCount desc

Sigma

title: Automated Non-Human User-Agent Bulk SaaS Data Exfiltration — UNC6671
id: 4b1e9d3f-1f6a-2c9e-0a01-8b7c2c9e0a05
status: experimental
description: Detects high-volume file access/download from Microsoft 365/SharePoint or Okta-connected SaaS via non-human user agents (python-requests, PowerShell, anomalous Okta mobile clients) or from known UNC6671 egress infrastructure, indicating scripted bulk exfiltration following account takeover.
references:
    - https://cloud.google.com/blog/topics/threat-intelligence/unc6671-targets-financial-services-and-enterprise-cloud-environments
    - https://cloud.google.com/security/report/resources/cloud-threat-horizons-report-h1-2026
author: Threadlinqs Team
date: 2026-08-09
tags:
    - attack.discovery
    - attack.t1087
    - attack.t1526
    - attack.exfiltration
    - attack.t1537
    - attack.t1020
    - attack.command-and-control
    - attack.t1071
    - attack.group.unc6671
logsource:
    product: m365
    service: threat_management
detection:
    selection_ua:
        UserAgent|contains:
            - 'python-requests'
            - 'PowerShell'
    selection_okta_ua:
        UserAgent: '0811A9866E.com.okta.android.auth/8.18.0'
    selection_ip:
        ClientIP:
            - '23.234.75.84'
            - '195.140.213.114'
            - '195.140.213.115'
    selection_op:
        Operation:
            - 'FileDownloaded'
            - 'FileAccessed'
            - 'FileSyncDownloadedFull'
            - 'SearchExported'
    condition: selection_op and (selection_ua or selection_okta_ua or selection_ip)
falsepositives:
    - Legitimate automation/RPA service accounts or backup jobs using python-requests or PowerShell against Graph/SharePoint APIs — allowlist known service-account UPNs and their registered app IDs
    - Authorized data-loss-prevention or eDiscovery bulk exports (SearchExported by compliance admins)
level: high

Known false positives:

  • Legitimate RPA/service-account automation using python-requests or PowerShell against Graph/SharePoint APIs — allowlist known service-account UPNs and registered app IDs
  • Authorized eDiscovery/compliance bulk exports

Tuning notes: Tune the count/distinct_files thresholds (50 events / 20 files per hour) to the tenant's baseline service-account behavior; add a lookup of approved automation UPNs to auto-suppress known-good scripted access before alerting on the raw UA match.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • AiTM Passkey/SSO Phishing Portal Access Following Vishing Pretext (UNC6671) — Sentinel · unlock with TL Creds
  • Fraudulent Passkey/MFA Enrollment with Security Notification Suppression (UNC6671 Account Takeover) — Sentinel · unlock with TL Creds
  • Automated Non-Human SaaS Bulk Data Exfiltration to External Egress Infrastructure (UNC6671) — Sentinel · unlock with TL Creds
  • AiTM Passkey/SSO Phishing Portal Access Following Vishing Pretext (UNC6671) — Sigma · unlock with TL Creds
  • Fraudulent Passkey/MFA Enrollment with Security Notification Suppression (UNC6671 Account Takeover) — Sigma · unlock with TL Creds
  • Automated Non-Human SaaS Bulk Data Exfiltration to External Egress Infrastructure (UNC6671) — Sigma · unlock with TL Creds
  • AiTM Passkey/SSO Phishing Portal Access Following Vishing Pretext (UNC6671) · unlock with TL Creds
  • Fraudulent Passkey/MFA Enrollment with Security Notification Suppression (UNC6671 Account Takeover) · unlock with TL Creds

Indicators of compromise

As of 2026-08-09, Threadlinqs Intelligence tracks 25 indicators of compromise for TL-2026-1963.

Sample indicators (3 of 25)

  • IP address: 31.7.56[.]61
  • Domain: activatepasskey[.]com
  • IP address: 31.7.56[.]52

Defanged, TLP:CLEAR. Full set of 25 indicators: unlock with TL Creds.

25 indicators of compromise · unlock with TL Creds. Compare plans

Community OSINT corroboration

14 of this threat's indicators have also been reported by the open-source security community, which observed at least one of them before this report was published. Community sightings are unverified and are kept separate from Threadlinqs' curated indicators. Indicator values, reporters and campaign linkage are available to authenticated Red-tier users.

Technical analysis

Point72 confirmed the attack and found no client data stolen; Two Sigma said it blocked the attempt outright. Security reporting ties the TTPs and July 2026 sector-targeting pivot to UNC6671, the multi-brand data-theft extortion group behind the BlackFile/Redact/Pink/Helix/Falcon brands.

Beginning August 5, 2026, four of Wall Street's largest hedge funds — Point72 Asset Management, Citadel, Two Sigma Investments, and Millennium Management — along with several unnamed private-equity firms were targeted in a coordinated voice-phishing (vishing) campaign that used AI voice-cloning to impersonate trusted IT support staff and colleagues. Attackers called employees and IT help-desk agents directly (in some cases dialing personal mobile numbers rather than office lines to bypass corporate call-monitoring controls), used a cloned voice replicating a trusted colleague's or IT technician's tone and phrasing, and pressured the target into an urgent 'security migration' or 'passkey enrollment' pretext — asking the help-desk agent to reset MFA/credentials or enroll a new authentication device in real time.

Point72 notified investors it had been hit and that its initial review found no client data had been stolen, though the investigation was continuing; the firm declined further public comment. Two Sigma stated it successfully blocked the attempt with no impact to data or systems. Citadel and Millennium Management declined to comment on whether their defenses were breached. FINRA confirmed its Financial Intelligence Fusion Center (FIFC) — a near-real-time threat-intelligence-sharing portal for member firms launched March 31, 2026 — was activated and in contact with affected firms, marking the FIFC's first live operational test since launch.

While the source reporting on Point72 specifically noted 'Scattered Spider' only as a group known for similar social-engineering tactics historically (against targets such as Erie Insurance and Aflac) without confirming direct attribution, subsequent technical reporting (BleepingComputer, Google Threat Intelligence Group/Mandiant, The Hacker News) ties the TTPs and timing of this campaign to UNC6671 — a financially motivated data-theft extortion cluster that has operated under the brand names BlackFile (emerged February 2025, retired/disputed-shutdown May 11, 2026), Redact (rebrand announced June 27, 2026), and the concurrently active Pink, Helix, and Falcon brands. GTIG assesses a single core intrusion group likely drives the helpdesk-vishing and cloud-data-theft activity behind these brands, based on shared phishing-panel infrastructure, code/template reuse, and continued Bitcoin cashouts through the disputed 'shutdown' period. UNC6671's targeting shifted from manufacturing/healthcare/real-estate/technology/transportation/hospitality sectors toward private equity, hedge funds, major law firms, and financial-rating agencies starting July 2026 — directly preceding the August 5 hedge-fund wave.

UNC6671's documented kill chain: vishing calls to personal mobile devices posing as corporate IT help desk mandating urgent FIDO2 passkey enrollment or MFA changes; victims/agents directed to adversary-in-the-middle (AiTM) phishing portals on lookalike 'passkey'/'sso' domains that harvest credentials and session/MFA tokens in real time; attacker-controlled MFA devices registered on the compromised account while legitimate MFA and password-reset notification emails are deleted to evade detection; the compromised Okta/Microsoft Entra ID SSO account is then used as a single point of entry into connected SaaS platforms (Microsoft 365, Google Workspace, Salesforce and its OAuth-connected ecosystem); automated, script-driven bulk data exfiltration follows (identifiable by non-human user-agent strings hitting cloud APIs), after which the group extorts victims via brand-specific data-leak sites, with initial demands of $1-3M USD typically settling around $750K (a 50-75% reduction) in the majority of tracked cases. Between January 7 and May 12, 2026 alone, GTIG tracked ~141.65 BTC (~$10.6-10.7M USD) in ransom payments across 18 wallets tied to the BlackFile era of the operation.

MITRE ATT&CK techniques by tactic:

Exfiltration

T1020 Automated Exfiltration; T1537 Transfer Data to Cloud Account

Command and Control

T1071 Application Layer Protocol

Discovery

T1087 Account Discovery; T1526 Cloud Service Discovery

Persistence

T1098 Account Manipulation

Credential Access

T1110 Brute Force; T1557 Adversary-in-the-Middle

Impact

T1531 Account Access Removal

lateral-movement

T1550 Use Alternate Authentication Material

defense-impairment

T1556 Modify Authentication Process

Initial Access

T1566 Phishing

Reconnaissance

T1589 Gather Victim Identity Information; T1598 Phishing for Information

Remediation

Immediate actions

  • Alert help-desk/IT support staff to the active pretext (urgent FIDO2 passkey enrollment or MFA-reset requests arriving via voice call, especially to personal mobile numbers) and require out-of-band, non-phone-based identity verification before any credential or MFA action
  • Review Okta/Entra ID logs for 'system.multifactor.factor.setup' events preceded by authentication failures, and for MFA device enrollments/resets performed shortly after help-desk contact
  • Search SaaS audit logs (SharePoint/M365 FileAccessed, Okta System Log) for high-volume file access or auth events from non-human user agents (python-requests, PowerShell, generic Go/Okta-mobile clients) or from residential-proxy/commercial-VPN source IPs inconsistent with employee baselines
  • Block/flag the known UNC6671 lookalike phishing domains and proxy/exfiltration IPs listed in this record at email/web gateways and SIEM watchlists

Workarounds

  • Require call-back verification to a pre-registered corporate extension (never a caller-supplied number) before any help-desk-initiated credential or MFA change
  • Temporarily restrict help-desk MFA-reset authority to require secondary manager/security-team approval for financial-sector and other high-value target roles

Longer-term hardening

  • Deploy phishing-resistant MFA (FIDO2 hardware security keys, platform passkeys, Windows Hello) — WebAuthn's cryptographic origin binding defeats AiTM token relay
  • Enforce centralized SSO with conditional-access policies restricting authentication to corporate-managed devices/networks (MDM/EDR-enrolled, VPN/SASE), blocking unmanaged personal-device logins
  • Reduce session lifetimes and require re-authentication at least once per workday plus step-up auth for sensitive actions (MFA re-enrollment, password reset, high-privilege app access)
  • Remove human judgment from identity-verification decisions at the help desk — replace phone-based 'trust your gut' verification with automated, policy-enforced verification (managed-device attestation, ticket/manager callback via pre-registered numbers)

Affected and fixed versions

  • N/A (social-engineering/identity-process attack, not a software vulnerability) — Okta / Microsoft Entra ID SSO, Microsoft 365, Google Workspace, and connected SaaS/OAuth-integrated applications (e.g. Salesforce ecosystem)
    Vulnerable versions: Any tenant relying on phone-verifiable or knowledge-based help-desk identity verification for MFA/credential reset
    Fixed in: Tenants enforcing FIDO2/WebAuthn phishing-resistant MFA with managed-device conditional access

Timeline

  • BlackFile extortion brand first emerges, initially targeting retail and hospitality sectors
  • GTIG begins tracking BlackFile-era Bitcoin ransom payment window (through May 12, 2026), ultimately totaling ~141.65 BTC (~$10.6-10.7M USD) across 18 wallets
  • AiTM phishing domains locked-ioc-1 and locked-ioc-3 registered via TUCOWS, fronted by Cloudflare, targeting financial services/hospitality/energy
  • Bridge domain locked-ioc-4 registered, later observed linking BlackFile-branded to Pink-branded victim campaigns
  • BlackFile brand shutdown publicly announced; Bitcoin cashouts to BlackFile wallets continue afterward, contradicting the announced shutdown
  • AiTM domains locked-ioc-12 and locked-ioc-13 registered (NICENIC/EZYDOMAIN) for Helix-branded targeting of financial services and legal sector victims
  • Redact extortion brand publicly launches, describing itself as a rebrand necessitated by a compromised/rogue BlackFile affiliate
  • UNC6671 targeting pivots from manufacturing/healthcare/real-estate/technology/transportation/hospitality toward private equity, hedge funds, major law firms, and financial-rating agencies
  • locked-ioc-2 registered (NICENIC, Cloudflare/Private Layer), later observed serving both Falcon- and Helix-branded victim campaigns simultaneously — evidence of shared operator infrastructure
  • Coordinated AI voice-cloning vishing attacks strike Point72, Citadel, Two Sigma, and Millennium Management; Two Sigma blocks the attempt outright; FINRA's Financial Intelligence Fusion Center is activated for affected member firms in its first live operational test
  • Fortune, TechTimes, InvestmentNews, TechNadu, and Benzinga report on the coordinated hedge-fund vishing wave and FINRA FIFC activation
  • BleepingComputer and other outlets report the hedge-fund attack wave's TTPs and timing align with UNC6671 (BlackFile/Redact/Pink/Helix/Falcon), based on GTIG's July 2026 sector-targeting shift and documented AiTM helpdesk-vishing kill chain
  • Point72 formally notifies investors of the incident, stating its initial review found no client information had been stolen, and declines further public comment; The Cyber Express publishes coverage citing possible Scattered Spider TTP overlap without confirming attribution

Sources cited

More from UNC6671

Previous and next in phishing

Drafted by the Threadlinqs research pipeline from 15 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats