CTIPilot
AI-generated · no human review · verify critical claims against the linked source. how it works →
‹Sat · 30 May 2026›
All daily briefs →
Daily brief · UTC day

Saturday, 30 May 2026

12 verified findings from 1 run · the settled record for this UTC day, in the classic brief order.

Criticality
Kind
Topic
Region
TL;DR · the day in one read
  1. 01ESET APT Activity Report Q4 2025–Q1 2026: Sandworm strikes NATO energy, Lazarus targets EU drone sector, UNC5221 pivots to Ivanti SPAWN toolset. ESET APT Activity Report Q4 2025–Q1 2026: Sandworm wiper targets Polish NATO energy company; Lazarus targets European drone manufacturers; UNC5221 deploys a new SPAWN toolset implant against Ivanti VPN appliances (ESET WeLiveSecurity, 2026-05-28). →
  2. 02CVE-2026-48710 "BadHost", Starlette (FastAPI / vLLM / LiteLLM / MCP SDK): Pre-Auth Auth Bypass via Malformed Host Header. CVE-2026-48710 "BadHost", Starlette/FastAPI host-header auth bypass hits AI/ML serving infrastructure including vLLM, LiteLLM, and MCP servers (NCSC-NL NCSC-2026-0171, 2026-05-29). A single malformed Host header character shifts request.url.path so middleware grants access to an unintended route. Fix: Starlette ≥ 1.0.1. →
  3. 03Qilin and Inc ransomware exploit the GlobalProtect bypass; one intrusion pivoted to vCenter and ESXi past in-guest EDR. CVE-2026-0257 is a pre-auth VPN authentication bypass in PAN-OS GlobalProtect (CISA KEV; Palo Alto reports limited exploit attempts, and Rapid7 and Unit 42 report successful exploitation): an attacker forges valid auth-override cookies by re-using the GlobalProtect certificate shared with the HTTPS service, with no credentials. CISA's KEV catalog (version 2026.10.04) marks it as used in known ransomware campaigns, and Arctic Wolf's report of 2026-07-20 describes June 2026 intrusions in which Qilin ransomware deployment began with this bypass. Unit 42 now names Qilin and Inc as exploiting it and describes one intrusion that pivoted from the VPN gateway to vCenter and ESXi and stole domain credentials offline from a domain controller's virtual disk, which in-guest EDR does not see. Patch or disable auth-override cookies, and treat a gateway that was exposed and unpatched as a compromise-assessment target that includes the virtualization and backup tier. →
  4. 04Ghost Stadium PhaaS, 300+ FIFA domain clones, multi-language fake SSO, targeting UK/Germany/Portugal/Spain fan credentials before June 11 kickoff. Ghost Stadium PhaaS, 300+ pixel-perfect FIFA domain clones targeting UK, Germany, Portugal, Spain fan credentials ahead of 11 June kickoff (FBI IC3 PSA260527, 2026-05-27); Chinese-speaking operator running multi-language fake SSO. →

01Active threats, incidents & disclosures4 items

HIGH

Ghost Stadium PhaaS, 300+ FIFA domain clones, multi-language fake SSO, targeting UK/Germany/Portugal/Spain fan credentials before June 11 kickoff

The FBI issued PSA260527 on 27 May 2026 warning that a Chinese-speaking financially-motivated threat actor tracked by Group-IB as Ghost Stadium has deployed more than 300 phishing sites impersonating fifa.com, all reproducing the official site pixel-for-pixel including a fake single-sign-on authentication flow in multiple languages (FBI IC3 PSA260527, 2026-05-27; BleepingComputer, 2026-05-28). Typosquatted domains span alternative TLDs (.org, .xyz, .live, .sale) and character substitutions; additional fake employment portals impersonate FIFA HR functions. Criminal objectives include credential and financial-data theft via the fake SSO, counterfeit ticket and hospitality sales, fake merchandise and streaming-rights fraud. UK, Germany, Portugal, and Spain are explicitly named as target demographics. Browser-based security controls (Safe Browsing, SmartScreen) do not protect against freshly-registered domains before abuse is reported. For defenders at organisations with large employee populations purchasing World Cup tickets: advise bookmarking https://www.fifa.com directly; treat any search-result-sponsored result for FIFA ticket purchases as unverified. The high-intensity fraud window is the lead-up to the July 19 final.

threat30 May 05:00Zmulti-sourceOpen finding →
NOTABLE

GREYVIBE, newly documented Russia-nexus cluster deploys five parallel attack chains against Ukraine with AI-generated lures and two PowerShell RATs

WithSecure Labs disclosed GREYVIBE on 28–29 May 2026, a previously-unnamed Russia-nexus threat cluster active since at least August 2025, targeting Ukrainian military, government, civilians, and businesses (WithSecure Labs, 2026-05-29; SecurityWeek, 2026-05-28). Five parallel attack chains: PhantomMail (spear-phishing with ZIP/RAR archives via Google Drive and 4sync), PhantomClick (fake CAPTCHA/ClickFix pages impersonating Zoom and LAPAS), PrincessClub (fraudulent adult-club sites with WebRTC-based social engineering), DroneLink (counterfeit Ukrainian Armed Forces charity sites), and Nebo (fake Russian military login portals). Core malware: LegionRelay (PowerShell RAT with file theft, screenshots, credential harvesting, RDP access; RC4 C2 comms), PhantomRelay (PowerShell RAT with dynamic script loading and watchdog persistence), and FallSpy (Android spyware for contact, call log, and geolocation extraction). Four custom obfuscators (LOOKVALPS, LOOKVALJS, DAYLIGHT, TEASOUP) were assessed as LLM-assisted developments. Attribution evidence: Russian-language panels and code comments; C2 servers in UTC+3 (Moscow time); OPSEC failures including public scan-platform uploads. WithSecure identifies possible links to UAC-0098 (former TrickBot associates). MITRE ATT&CK: T1566.001/T1566.002, T1059.001, T1005, T1204.001, T1133. Detection: alert on PowerShell spawned from archive-extraction utility parent processes; hunt scheduled tasks created by PowerShell beaconing to dynamic DNS; Android MDM alerts on sideloaded APKs accessing mic/camera. Organisations supporting Ukrainian government or civil-society counterparts are within the targeting scope.

threat30 May 05:00Zmulti-sourceOpen finding →
NOTABLE

CNIL fines IQVIA Operations France €5M for health data warehouse security failures: no MFA, no log monitoring, no network segmentation

France's CNIL fined IQVIA Operations France €5 million on 26 May 2026 for systematic GDPR violations across two authorised health data warehouses, LRX (fed by ~14,000 pharmacies) and EMR (fed by thousands of GPs) (CNIL, 2026-05-28). The CNIL enumerated five control failures: (1) IQVIA operated the warehouses outside the scope of its CNIL authorizations, deliberations 2018-289 and 2021-015 approved specific study types, and IQVIA conducted studies beyond those terms (Art. 66 of the French Data Protection Act); (2) patients were not informed that IQVIA acted as a data controller for their prescription data, violating GDPR Art. 14 information obligations; (3) multi-factor authentication was absent from all warehouse access paths; (4) no automated connection-log monitoring or alerting was in place, IQVIA confirmed retrospective deployment only after the CNIL investigation commenced; (5) no network segmentation between the health data warehouse and other IQVIA corporate infrastructure. The fine magnitude reflects the scope ("several tens of millions" of individuals) and IQVIA's market position. A compliance order with a €10,000/day penalty period accompanies the fine. For defenders this ruling operationalises baseline controls now explicitly expected for health data warehouse operations: MFA on all warehouse access paths, automated log alerting, network segmentation between sensitive-data stores and corporate infrastructure, and strict compliance with CNIL authorization scope for each study type conducted.

incident30 May 05:00Zmulti-sourceOpen finding →
NOTABLE

LLMShare malvertising campaign: attackers embed fake outage pages in ChatGPT share links and serve infostealer downloads via Google Ads

Push Security documented LLMShare, a malvertising campaign in which attackers buy Google Ads targeting "ChatGPT" and "ChatGPT download" queries (Push Security, 2026-05-29; BleepingComputer, 2026-05-29). Victims clicking the ads land on legitimate chatgpt.com/s/[unique-id] share URLs that render attacker-controlled HTML (a fake high-traffic outage page with a "Download our desktop app to continue" button) directly from the OpenAI domain. Because chatgpt.com is trusted by enterprise web-filtering rules and firewalls, the landing page is not blocked. The download button redirects to an attacker-controlled domain impersonating OpenAI; the site uses cloaking (serves a benign page to scanners). Windows users receive an infostealer payload. The technique exploits the same ChatGPT Artifacts/sharing feature previously abused in the ACR Stealer campaign (covered 2026-05-26) and extends it to malvertising. Detection: monitor for browser-spawned executable downloads from chatgpt.com domains; legitimate ChatGPT desktop app downloads do not originate from that path; alert on unusual process launch from browser-extracted or browser-downloaded unsigned executables. MITRE ATT&CK: T1566.002, T1204.001, T1036, T1027.

threat30 May 05:00Zmulti-sourceOpen finding →
HIGHCVE-2026-0257exploitedupdatedNATOA1

CVE-2026-0257, Palo Alto PAN-OS GlobalProtect: pre-auth authentication bypass via certificate reuse, marked by CISA as used in ransomware campaigns

Authentication override cookies in PAN-OS GlobalProtect let a portal or gateway issue a cookie that a user presents in place of re-authenticating, and the feature is not enabled by default (Rapid7, 2026-05-29). Palo Alto Networks says the flaw affects firewalls with a GlobalProtect portal or gateway where authentication override cookies are enabled and a specific certificate configuration exists, classes it as CWE-565 (reliance on cookies without validation and integrity checking), and scores it CVSS 4.0 7.8 with exploit maturity Attacked (Palo Alto Networks PSIRT, 2026-05-13). Rapid7's analysis is that when the certificate used for the cookies is reused for another feature such as the HTTPS service of the portal or gateway, a remote unauthenticated attacker can discover that certificate's public key, forge an arbitrary cookie that the server decrypts and trusts, and bypass authentication; Rapid7 Labs published a proof-of-concept script that tests for it (Rapid7, 2026-05-29). Palo Alto Networks says it is aware of limited exploit attempts on unpatched devices without mitigations applied (Palo Alto Networks PSIRT, 2026-05-13); Rapid7 MDR dates the earliest observed exploitation to 17 May 2026 and saw two waves, on 18 May from Vultr-hosted infrastructure and on 21 May from Dromatics Systems, which it believes are likely the same threat actor because of a consistent MAC address (Rapid7, 2026-05-29). CISA's KEV catalog lists the CVE with a date added of 29 May 2026 (CISA KEV catalog, 2026-10-04).

Affected are PAN-OS 12.1, 11.2, 11.1 and 10.2 releases below the fixed builds in Palo Alto's table and Prisma Access 11.2 and 10.2; Panorama and Cloud NGFW are not affected (Palo Alto Networks PSIRT, 2026-05-13). The solution table names the first fixed build for each minor release, for example 12.1.4-h6 or 12.1.7 and later for 12.1.2 through 12.1.4, 11.2.12 or later for 11.2.11, 11.1.15 or later for 11.1.14, and 10.2.18-h6 or later on the 10.2 train, while Prisma Access 11.2 needs 11.2.7-h13 or later and 10.2 needs 10.2.10-h36 or later (Palo Alto Networks PSIRT, 2026-05-13). After the upgrade users must re-authenticate once, because the firewall regenerates the cookie with a more secure method (Palo Alto Networks PSIRT, 2026-05-13). Palo Alto's mitigations are a dedicated certificate used only for authentication override cookies, or disabling Authentication Override (Palo Alto Networks PSIRT, 2026-05-13).

Triage: cookie logins are routine wherever Authentication Override is enabled, because users present the cookie in lieu of re-authenticating (Rapid7, 2026-05-29). In some intrusions Arctic Wolf saw a cookie-based authentication failure with the error Cannot decrypt cookie, followed quickly by a successful cookie-based authentication from the same source, although the error was not present in every intrusion (Arctic Wolf, 2026-06-11).

Palo Alto Networks has become aware of limited exploit attempts on unpatched PAN-OS devices without mitigations applied.

Palo Alto Networks PSIRT

Due to the consistent MAC address, Rapid7 believes both waves of exploitation are likely from the same threat actor (TA).

Rapid7

Arctic Wolf investigated multiple distinct intrusions during June 2026 that resulted in Qilin ransomware deployment, all originating from exploitation of CVE-2026-0257 against Palo Alto Networks firewall appliances.

Arctic Wolf Labs 2026-07-20

Ransomware operators exploiting the May 2026 CVE-2026-0257 include Spikey Scorpius (aka Qilin) and Tarnished Scorpius (aka Inc).

The most significant evolution in tactics, techniques and procedures (TTPs) following successful exploitation of CVE-2026-0257 is the direct pivot to the virtualization layer.

This tactic bypasses in-guest endpoint detection and response (EDR) and security controls, representing a highly sophisticated method for obtaining domain-wide credentials.

Unit 42
Updaterun 2026-06-10-c84347b2actionsevidencesectorssourcesbody

Unit 42's 9 June update on CVE-2026-0257 confirms that a limited number of probed PAN-OS GlobalProtect devices had attacker-established, gateway-connected VPN sessions, moving this from "exploit attempts observed" to confirmed successful exploitation (Unit 42, 2026-06-09). The bug is CWE-565, reliance on cookies without validation and integrity checking (Palo Alto Networks PSIRT, 2026-05-13); when the cookie certificate is shared with another function, an attacker can discover its public key and forge authentication-override cookies, and Rapid7 dates successful exploitation to 17 May from low-cost hosting IPs (Rapid7, 2026-05-29).

Affected: PAN-OS 10.2/11.1/11.2/12.1 and Prisma Access where authentication override is enabled with a shared certificate; patched in 12.1.7+, 11.2.12+, 11.1.15+, 10.2.18-h6+ and corresponding Prisma builds (Palo Alto Networks, 2026-05-13). After the upgrade users must re-authenticate once, because the fixed firewall regenerates the cookie with a more secure method; as a workaround disable authentication override or assign it a dedicated certificate (Palo Alto Networks, 2026-05-13). Hunt GlobalProtect gateway logs for cookie-based authentications from unexpected source addresses; Rapid7's log of the attack shows the authentication method recorded as a cookie (Rapid7, 2026-05-29).

Updaterun 2026-06-17-e102009cevidenceregionssectorssourcesbody

Palo Alto's Unit 42 confirms an active exploitation campaign against the GlobalProtect cookie authentication-bypass (CVE-2026-0257) (Unit 42, 2026-06-09). The flaw (CWE-565, reliance on cookies without validation and integrity checking) lets an attacker forge a session and establish a VPN tunnel without credentials when the override feature is enabled (Palo Alto Networks PSIRT).

Arctic Wolf saw exploitation attempts across insurance, finance, manufacturing, education, engineering and healthcare targets in North America and Europe, and in a subset of intrusions automated internal SMB reconnaissance consistent with Impacket tooling (anonymous NTLM logon, share enumeration and domain-user discovery); it says the follow-on activity did not progress substantially beyond the initial SMB session setup and lightweight reconnaissance (Arctic Wolf, 2026-06-11). NCSC-CH refreshed its Security Hub advisory on 2026-06-16 to flag the Unit 42 confirmation (NCSC-CH Security Hub, 2026-06-16). Defenders: disable "Authentication Override" if not required, patch to fixed PAN-OS builds, and audit sessions since 17 May for anonymous NTLM logons and SMB share enumeration from VPN-assigned addresses.

Updaterun 2026-10-06T0405Z-inteltitleheadlinesummarypriorityimmediate_actiontagsentitiestechniquesaffected_productssourcesevidencesourcing_noteclassificationactionsbody

CISA's Known Exploited Vulnerabilities catalog (version 2026.10.04) now marks CVE-2026-0257 as used in known ransomware campaigns; the record names no group and its date of addition remains 2026-05-29 (CISA KEV catalog, 2026-10-04). Arctic Wolf Labs' report of 2026-07-20 says it investigated multiple distinct intrusions in June 2026 that ended in Qilin ransomware deployment, all originating from exploitation of CVE-2026-0257 against Palo Alto Networks firewalls, and that the post-exploitation tradecraft ranged from rapid encryption-only operations to double extortion, possibly suggesting several affiliates of the Qilin ransomware-as-a-service operation (Arctic Wolf, 2026-07-20).

The chain Arctic Wolf describes starts with a GlobalProtect VPN session obtained through the cookie bypass, in some cases from systems that identified themselves as kali, and continues with registry Run-key persistence that points at a ransomware payload staged under C:\PerfLogs, remote-access tools (AnyDesk, Ngrok, LogMeIn and, in one case, a scheduled task consistent with MeshAgent), LSASS credential theft through rundll32 and comsvcs.dll into a file with an .odt extension, a full copy of the Active Directory database with ntdsutil, lateral movement with PsExec over administrative shares and RDP, a PowerShell routine that clears every Windows event log with records, and Microsoft Defender real-time protection disabled in some cases (Arctic Wolf, 2026-07-20). In the double-extortion cases data went to the MEGA cloud storage service with Rclone before encryption, and the attackers targeted the Veeam backup infrastructure before deploying the ransomware; some intrusions involved no data theft before encryption (Arctic Wolf, 2026-07-20). A gateway that was reachable and unpatched therefore needs a compromise assessment of the Windows estate behind it, not only the patch.

Correctionrun 2026-10-11T1307Z-auditbodyevent_date

Arctic Wolf's June telemetry shows automated SMB reconnaissance consistent with Impacket tooling (anonymous NTLM logon, share enumeration, domain-user discovery) in a subset of intrusions, and Arctic Wolf states the activity did not progress substantially beyond the initial SMB session setup and lightweight reconnaissance (Arctic Wolf, 2026-06-11); the 2026-06-17 note had called it lateral movement. Before the Qilin reporting, Rapid7 saw no indication of successful lateral movement from the compromised devices (Rapid7, 2026-05-29), and Unit 42 reported no post-access behavior or lateral movement as of 9 June (Unit 42, 2026-06-09). Rapid7 saw the same spoofed MAC address on the attacker's client in both May waves, a pivot for VPN session logs (Rapid7, 2026-05-29), although Arctic Wolf found MAC spoofing was not consistent across later intrusions (Arctic Wolf, 2026-06-11). Rapid7 Labs published a public proof of concept on 29 May, the same day CISA added the flaw to its catalog (Arctic Wolf, 2026-06-11).

Updaterun 2026-10-21T0256Z-intelheadlinesummaryentitiessourcesevidencesourcing_noteactionsbody

Unit 42 updated its threat brief on 2026-10-19 and 2026-10-20: it continues to see attempted exploitation by ransomware operators, naming Spikey Scorpius (Qilin) and Tarnished Scorpius (Inc), and says it is aware of public reporting that attributes exploitation to actors using Settra ransomware (Unit 42, 2026-10-20). Its investigations have run since early June 2026 and span industries including government (Unit 42, 2026-10-20). The attackers use the flaw for an initial foothold, bypass multi-factor authentication to establish a VPN session, then run reconnaissance with dual-use remote monitoring and management tools, move over RDP and SMB and stage credential harvesting, exfiltration and ransomware (Unit 42, 2026-10-20).

The change Unit 42 stresses is a direct pivot to the virtualization layer: in one case the attackers went from the compromised VPN gateway to the vCenter management appliance, disabled the execInstalledOnly security control on ESXi hosts and used a custom tool for offline credential dumping from a domain controller's virtual disk, which bypasses in-guest EDR; it also saw this access chained with exploit tools aimed at Veeam backup infrastructure (Unit 42, 2026-10-20). The brief adds that GlobalProtect profiles mapped to HTTP or HTTPS services on management interfaces can expose the management interface on ports 4443 and 3879 without the administrators knowing, and that an out-of-band HTTPS connection to the management address on those ports shows whether it is exposed (Unit 42, 2026-10-20). The brief dates the flaw's addition to CISA's catalog to 13 May 2026, whereas the catalog's own record says 29 May 2026 (Unit 42, 2026-10-20; CISA KEV catalog, 2026-10-04).

Correctionrun 2026-10-21T0800Z-auditactionscvesbody

The one-time re-authentication is a consequence of the upgrade, not a step to force: with the fix, a firewall that uses authentication override cookies regenerates the cookie with a more secure method, so GlobalProtect users re-authenticate once after the upgrade even when they hold a valid cookie (Palo Alto Networks PSIRT, 2026-05-13). Cookie logins are routine where Authentication Override is enabled; in some intrusions Arctic Wolf saw a cookie-based authentication failure with the error Cannot decrypt cookie, followed quickly by a successful cookie-based authentication from the same source (Arctic Wolf, 2026-06-11).

Correctionrun 2026-10-25T1308Z-auditbodyevidence

The CWE-565 classification (reliance on cookies without validation and integrity checking) is Palo Alto Networks' (Palo Alto Networks PSIRT, 2026-05-13). Rapid7 describes the attacker discovering the public key of a certificate shared with another function and forging authentication-override cookies with it, and its proof of concept takes the key from the certificate chain the HTTPS service presents (Rapid7, 2026-05-29).

vulnerability30 May 05:00Zmulti-sourceOpen finding →

CVE-2026-48710 "BadHost", Starlette (FastAPI / vLLM / LiteLLM / MCP SDK): Pre-Auth Auth Bypass via Malformed Host Header

Starlette < 1.0.1 reconstructs request.url by concatenating the HTTP Host header with the request path and re-parsing the composite string, but validates each component under separate rules (X41 D-Sec Advisory x41-2026-002, 2026-05-22; GitHub Advisory GHSA-86qp-5c8j-p5mr). Injecting a /, ?, or # into the Host header (e.g. Host: example.com/health?x=) shifts the path boundary reported by request.url.path, causing middleware applying path-based access control to authorise access to an unintended route while the ASGI handler serves the attacker-specified one. A single curl -H 'Host: foo?' localhost:8000/admin bypasses authentication. Root cause: CWE-436 (Interpretation Conflict). CVSS 3.1 = 6.5 (GitHub Advisory); X41 scores 7.0 under CVSS 4.0. Affected: all Starlette versions ≥ 0.8.3, < 1.0.1; downstream dependents include FastAPI, vLLM, LiteLLM, Google ADK-Python, BentoML, Gradio, Langflow, Open WebUI, and the Python MCP SDK, approximately 325 million weekly downloads and 400,000+ GitHub dependents. Discovered by X41 D-Sec during an OSTIF-sponsored vLLM audit. Nginx, Apache, Caddy, Traefik, HAProxy, and Cloudflare all terminate malformed Host headers upstream; only direct-listen Python ASGI deployments without a compliant reverse proxy are exposed. No confirmed exploitation as of publication. NCSC-NL issued advisory NCSC-2026-0171 on 29 May; CCB Belgium issued a "Patch Immediately" advisory. Fix: upgrade Starlette to ≥ 1.0.1 (or pull FastAPI ≥ 0.115.5, vLLM ≥ 0.23.0, or the equivalent downstream package that pins the fixed Starlette). If patching is not immediately possible, place a compliant reverse proxy in front of any ASGI application using path-based access control. Detection: parse web-server access logs for Host header values containing /, ?, or # followed by path components.

CVE Summary Table

CVE Product CVSS EPSS KEV Exploited Patch Source
CVE-2026-0257 Palo Alto PAN-OS GlobalProtect 7.8 (CVSS 4.0) , Yes (2026-05-29) Yes, ITW waves 2026-05-18, 2026-05-21 10.2.7-h34+, 11.1.4-h33+, 11.2.4-h17+, 12.1.4-h6+ PAN PSIRT
CVE-2026-48710 Starlette / FastAPI / vLLM / LiteLLM / MCP SDK 6.5 (CVSS 3.1) / 7.0 (CVSS 4.0) , No No confirmed exploitation Starlette ≥ 1.0.1 GitHub Advisory
vulnerability30 May 05:00Zmulti-sourceOpen finding →
NOTABLECVE-2026-45585exploitedupdatedNATOA1

Microsoft calls Nightmare Eclipse's six uncoordinated Windows zero-day releases "never justifiable"; the three May drops were fixed in June 2026

In a blog post on 2026-05-27, the Microsoft Security Response Center said the vulnerabilities known as RedSun, UnDefend, BlueHammer, YellowKey, GreenPlasma and MiniPlasma "were not responsibly disclosed" and that uncoordinated disclosures putting proof-of-concept code for unpatched flaws "into the hands of bad actors are never justifiable"; it added that its Digital Crimes Unit "will continue bringing cases against these actors and those that enable their criminal activity" (Microsoft MSRC blog, 2026-05-27). The Record reported the post on 2026-05-29 as Microsoft's first response to the release campaign, noted that the researcher's GitHub account had since been removed, and reported that the researcher threatened a further release on 14 July 2026, the date of Microsoft's July Patch Tuesday (The Record, 2026-05-29).

Contradiction: the outlets read the statement differently. The Record says Microsoft "stopped short of directly threatening legal action" (The Record, 2026-05-29), while heise writes that the company "is threatening lawsuits and the police" (heise Security, 2026-05-29).

The first three drops, BlueHammer (CVE-2026-33825), UnDefend (CVE-2026-45498) and RedSun (CVE-2026-41091), all disclosed in April, have been exploited in live intrusions and appear on CISA's Known Exploited Vulnerabilities catalogue (The Record, 2026-05-29); UnDefend lets a standard user block Microsoft Defender definition updates (BleepingComputer, 2026-06-10). The three May drops had no patch and no confirmed exploitation when The Record reported (The Record, 2026-05-29). YellowKey, tracked as CVE-2026-45585, is a BitLocker security feature bypass that needs physical access (Microsoft MSRC). The researcher says MiniPlasma abuses the HsmOsBlockPlaceholderAccess routine of the Cloud Filter driver cldflt.sys, and that it is the flaw Google Project Zero reported in 2020 as CVE-2020-17103, "still present, unpatched"; BleepingComputer's test on a fully patched Windows 11 Pro system with the May 2026 updates opened a SYSTEM command prompt from a standard user account (BleepingComputer, 2026-05-17).

Microsoft fixed the three May drops on 2026-06-09: GreenPlasma as CVE-2026-45586, a CTFMON elevation of privilege that can give SYSTEM (MSRC CVE-2026-45586), MiniPlasma by re-releasing CVE-2020-17103 with the June 2026 updates (MSRC CVE-2020-17103), and YellowKey in the same Patch Tuesday (BleepingComputer, 2026-06-10).

Our Digital Crimes Unit will continue bringing cases against these actors and those that enable their criminal activity

Microsoft MSRC blog 2026-05-27

The company stopped short of directly threatening legal action

The researcher threatened a further release on July 14

The Record 2026-05-29

The company is threatening lawsuits and the police.

heise Security 2026-05-29

This CVE has been re-released to comprehensively address the vulnerability identified by CVE-2020-17103

MSRC CVE-2020-17103 2026-06-09
Updaterun 2026-10-21T1149Z-auditaffected_productsbodyclassificationcvesevent_dateevidenceheadlinereferencessourcessourcing_notesummarytagstechniquestitle

Microsoft fixed the three May drops in the June 2026 Patch Tuesday updates of 2026-06-09: YellowKey (CVE-2026-45585) and GreenPlasma, now tracked as CVE-2026-45586 (BleepingComputer, 2026-06-10), and MiniPlasma through CVE-2020-17103, which Microsoft re-released "to comprehensively address the vulnerability identified by CVE-2020-17103" (MSRC CVE-2020-17103). BleepingComputer also reports that Microsoft backtracked on its initial threats of legal action after blowback on social media, saying it would work with law enforcement when a security researcher "breaks the law and engages in malicious activity causing real harm to our customers" (BleepingComputer, 2026-06-10).

Builds on: Leaked BitLocker bypass "YellowKey" and CTFMON privilege escalation "GreenPlasma" were fixed in…

vulnerability30 May 05:00Zmulti-sourceOpen finding →

03Research, reports & policy5 items

HIGH

ESET APT Activity Report Q4 2025–Q1 2026: Sandworm strikes NATO energy, Lazarus targets EU drone sector, UNC5221 pivots to Ivanti SPAWN toolset

ESET published its APT Activity Report covering October 2025 through March 2026 on 28 May 2026 (ESET WeLiveSecurity, 2026-05-28). EU- and NATO-relevant findings for public-sector defenders: Sandworm (Russia/GRU) intensified destructive winter operations against Ukrainian infrastructure and targeted a Polish energy company in December 2025; a NATO member state critical-infrastructure attack attributed with medium confidence; this represents continued Sandworm willingness to conduct wiper operations beyond Ukraine's borders. Sednit/APT28 deployed Covenant and BeardShell implants against Ukrainian military, drone manufacturers, and logistics companies. Lazarus Group ran Operation DreamJob targeting European drone manufacturers, ESET assesses this as technology acquisition for North Korea's weapons programme. Operation DangerousPassword compromised the axios JavaScript library (100+ million weekly npm downloads), injecting trojanised code and demonstrating ongoing North Korea supply-chain interest in developer ecosystem targeting. UNC5221 (China-nexus) deployed a new implant assessed as part of the SPAWN toolset, specifically targeting Ivanti VPN appliances (Connect Secure, Policy Secure); organisations running unpatched Ivanti VPN should audit for SPAWN toolset artefacts including SPAWNANT installer, SPAWNMOLE tunneller, SPAWNSNAIL SSH backdoor, and SPAWNSLOTH log-tampering utility. The report PDF is available at https://web-assets.esetstatic.com/wls/en/papers/threat-reports/eset-apt-activity-report-q4-2025-q1-2026.pdf. Key defender actions: (a) confirm Sandworm wiper detection capability (file-destruction followed by MBR/VBR overwrite patterns, VSS deletion); (b) review Ivanti VPN logs for SPAWN footprints per CISA AA24-060A indicators; (c) audit npm dependency trees for axios versions <1.8.0 or 0.x released after the DangerousPassword campaign window.

annual-report30 May 05:00Zmulti-sourceOpen finding →
NOTABLE

Kimsuky (Velvet Chollima) deploys HTTPSpy RAT and Rust-based HelloDoor via VS Code Remote Tunnel and Cloudflare Quick Tunnel C2

ENKI WhiteHat and The Hacker News documented Kimsuky campaigns in March and April 2026 targeting South Korean military personnel and corporate entities with two malware chains (The Hacker News, 2026-05-29; ENKI WhiteHat, 2026-05-27). March chain: masquerade installers for nProtect Online Security and AhnLab Safe Transaction launch MemLoader.dll via regsvcs.exe, which downloads HTTPSpy. April chain: fake Webex meeting page delivers encrypted JavaScript (.jse extension) which stages a PowerShell downloader, ultimately installing HTTPSpy. HTTPSpy is a full-capability RAT (first observed 2022; previously used against a German defence manufacturer May–September 2024): RC4-encrypted C2, shell execution, file upload/download, screenshot capture, process injection, self-deletion. HelloDoor is a Rust-based PebbleDash variant (assessed LLM-assisted per ENKI): configurable sleep, command execution, directory traversal. C2 evasion: Kimsuky now abuses Visual Studio Code Remote Tunneling (authenticated via GitHub OAuth, registered via code --tunnel --name <name>) and Cloudflare Quick Tunnels (cloudflared.exe); neither can be blocked by IP or domain without blocking Microsoft and Cloudflare respectively. JSONPing confirms active infections via a locally-running HTTP server, reducing exposure of attacker infrastructure. MITRE ATT&CK: T1036 (Masquerading), T1059.001 (PowerShell), T1059.007 (JavaScript), T1071 (Application Layer Protocol). Detection: hunt for regsvcs.exe as a parent of DLL loads in non-.NET-Framework contexts; alert on VS Code CLI processes with --tunnel argument from non-developer endpoints; audit GitHub OAuth app grants for unrecognised VS Code tunnel registrations; monitor cloudflared.exe on managed endpoints without prior baseline.

research30 May 05:00Zmulti-sourceOpen finding →
NOTABLE

Red Canary: detecting Entra Agent ID privilege escalation, credential injection into agent blueprints enables lateral movement across the entire tenant

Red Canary published a detection-engineering primer on 27 May 2026 on the AgentIdentityBlueprint.AddRemoveCreds.All role in Microsoft Entra's new Agent ID identity class, autonomous app identities that act in a tenant without human interaction (Red Canary, 2026-05-27). A misconfigured or adversary-controlled agent identity holding this role can add client secrets to any agent blueprint, then authenticate as any agent identity in the tenant (including high-privilege ones) after legitimate credential rotation. The full privilege-escalation chain: agent app → malicious role assignment (AgentIdentityBlueprint.AddRemoveCreds.All) → credential injection into target blueprint → authenticate as high-privilege agent → pivot to all downstream resources that blueprint can access. Relevant log sources: AuditLogs, look for "Update application – Certificates and secrets management" with a non-human InitiatedBy.app.servicePrincipalId; MicrosoftGraphActivityLogs, Graph API calls from agent service principals with unusual IP and UserAgent fields; AADServicePrincipalSignInLogs, filter on Agent.agentType: agenticAppInstance. Correlation: match SignInActivityId from Graph logs to UniqueTokenIdentifier in sign-in logs to reconstruct credential-add-to-authentication chains. MITRE ATT&CK: T1098 (Account Manipulation), T1078.004 (Valid Accounts: Cloud Accounts). Swiss public-sector M365 deployments adopting AI agents via Copilot Studio or Azure AI Foundry should establish baselines for each agent identity's API scope and alert on credential additions to blueprints by any identity other than the provisioning pipeline. [SINGLE-SOURCE]

research30 May 05:00Zsingle-sourceOpen finding →
Sources: Red Canary
NOTABLE

ChatGPhish: Permiso Security documents ChatGPT Markdown renderer trusting third-party image URLs and links, used for IP exfiltration and phishing via legitimate chatgpt.com

Permiso Security's P0 Labs (researcher Andi Ahmeti) disclosed on 29 May 2026 that ChatGPT's web summarisation feature unconditionally trusts and renders Markdown image URLs and links extracted from third-party pages, executing them inside the trusted chatgpt.com UI (Permiso Security P0 Labs, 2026-05-29; The Hacker News, 2026-05-29). An attacker embedding a small Markdown payload on any web page (GitHub README, SaaS dashboard, documentation portal) triggers the attack when a victim asks ChatGPT to summarise the page: the payload executes silently and can exfiltrate the victim's IP, User-Agent, and Referer via attacker-hosted image fetch; render malicious links styled as ChatGPT output; inject fake security alerts; and serve QR codes from attacker-controlled S3 buckets that bypass desktop URL filters by moving the click action to mobile. Permiso submitted to OpenAI via Bugcrowd on 29 April; after follow-up on 7 May, OpenAI marked it as not reproducible then as not applicable, without resolution. No CVE assigned. Defenders using ChatGPT for document summarisation in enterprise workflows should: restrict ChatGPT access to internal documentation portals; educate users that any AI-summarised third-party page can carry attacker instructions embedded in rendered output.

research30 May 05:00Zmulti-sourceOpen finding →
NOTABLE

Sysdig TRT: first observed LLM-agent-driven post-exploitation, CVE-2026-39987 Marimo notebook RCE to database exfiltration in 4 pivots under one hour

Sysdig's Threat Research Team documented what they assess as the first in-the-wild LLM-agent-driven intrusion, observed on 10 May 2026 (Sysdig TRT, 2026-05-26; The Hacker News, 2026-05-29). Initial access: exploitation of CVE-2026-39987, a pre-auth RCE in Marimo notebook < 0.20.4 (patched in 0.23.0) on an internet-accessible instance (T1190). An AI agent then drove four autonomous pivots: (1) extracted two cloud credentials from the host filesystem (T1552.001); (2) replayed them via a Cloudflare Workers egress pool to call AWS Secrets Manager APIs and retrieve an SSH private key (T1555); (3) executed eight parallel SSH sessions against a downstream bastion (T1021.004); (4) exfiltrated the full schema and contents of a downstream PostgreSQL database within two minutes (T1048). Sysdig identified LLM agent involvement from four artefacts: improvised schema discovery without environmental foreknowledge; a Chinese-language planning comment in the command stream ("看还能做什么"; "See what else we can do"); machine-optimised command formatting (delimiter-separated, bounded output, stderr discarded, less disabled); and sequential hand-off of output values as inputs to subsequent commands. Cloudflare Workers obscured the origin IP. No attribution was made. Defender countermeasures: update Marimo to ≥ 0.23.0; restrict internet-accessible notebook deployments; monitor AWS CloudTrail for Secrets Manager GetSecretValue calls from unexpected IPs; restrict SSH bastion access to known CIDR ranges.

research30 May 05:00Zmulti-sourceOpen finding →

04Action items3 items

Verification & coverage notes1 run

2026-05-30-aca445cc · Claude Sonnet 4.6 · 14 entries published

Items dropped:

  • Dutch Asocks botnet takedown (S3), duplicate; covered in brief 2026-05-29 as item:dutch-police-ncsc-dismantle-asocks-residential-proxy-botnet.
  • Silent Ransom Group / Luna Moth in-person IT impersonation (S3, S4), duplicate; covered in brief 2026-05-28 as item:fbi-flash-csa-260526-silent-ransom-group-sends-operatives-ph.
  • Akira ransomware claims GS Yuasa Lithium Power (S4), dropped; single source (ransomware.live aggregator only), no victim disclosure, no HIGH-reliability journalistic corroboration. Per fake-news guard rule, dark-web listing claims require victim disclosure or HIGH-reliability journalism. Logged: Akira/GS Yuasa Lithium Power — single-source leak-site claim, no confirmation.
  • Pay Tel Azure misconfiguration exposure (S4), dropped; US-only incident, minimal operational relevance for CH/EU public-sector SOC; no regulatory action in window.
  • CVE-2026-8992 from § 2, dropped from trending vulnerabilities; does not clear any § 2 gate (no CISA KEV, no ENISA EUVD exploited=true, no vendor-confirmed ITW, local LPE only). Covered in § 4 UPDATE as NCSC.ch advisory delta.
  • CVE-2026-39987, not listed in § 2; no CISA KEV, no vendor advisory confirming exploitation (observation by Sysdig only), Marimo is not sufficiently widely-deployed to trigger the pre-auth-RCE gate. Covered in § 3 research context.

Single-source items:

  • ENISA NIS360 2026 (§ 3); ENISA is the primary disclosing authority for its own report; national-CERT carve-out applies. PD-5 satisfied.
  • CNIL IQVIA €5M fine (§ 1); CNIL is the primary regulator; single-source acceptable under national-CERT carve-out. PD-5 satisfied.
  • Red Canary Entra Agent ID primer (§ 3), marked [SINGLE-SOURCE] in body; Red Canary is a HIGH-reliability research publisher; no threat-actor involvement (detection-engineering article).
  • Ivanti SAC CVE-2026-8992 UPDATE (§ 4), NCSC Switzerland updated advisory; NCSC.ch is primary disclosing authority for CH-jurisdiction advisories; carve-out applies.

Recency notes:

  • Sysdig TRT Marimo intrusion primary source dated 2026-05-26 (52 hours before window start at ~2026-05-28 16:00 UTC); included on the basis of in-window THN amplification (2026-05-29) and operational novelty (first observed LLM-agent-driven intrusion). Flagged per PD-7 recency re-check.

Sub-agent telemetry:

  • S1: Claude Sonnet 4.6, 518s, 5 items
  • S2: Claude Sonnet 4.6, 484s, 5 items
  • S3: Claude Sonnet 4.6, 614s, 8 items (2 deduped)
  • S4: Claude Sonnet 4.6, ended_at=04:24:35Z vs started_at=04:12:33Z → 722s actual wall-clock; YAML duration_seconds=3927 is a sub-agent computation error; actual run time 12 min, within cap

Coverage gaps: inside-it-ch (Cloudflare 403 ×3); ncsc-ch-security-hub (bridged, no new in-window items beyond covered advisories); sophos-xops (HTTP 503 ×2 on both feed URLs, no fallback content); databreaches-net (persistent 403 ×5 runs); cert-fr-avis (no in-window advisory); cert-eu (no in-window advisory); sec-edgar-8k (0 hits in window); edpb (no in-window notices); ico-uk (no in-window enforcement in window).

Verification: Phase 5.7 complete, 5 iterations (cap reached). All findings remediated. Iter 1 (Opus): NEEDS_FIXES (truth: 2, advisory: 3), IOC hard-rule violations (MAC + defanged domain), "Beagle" cross-campaign name, World Cup date, MSRC misanchor, CWE-444. Iter 2 (Sonnet): NEEDS_FIXES (truth: 1), GDPR Art. 21 not in CNIL source; correct framing is Art. 66 French DPA scope violation. Iter 3 (Opus): NEEDS_FIXES (truth: 2, advisory: 4), unsourced implant name removed; MAC descriptor corrected. Iter 4 (Sonnet): NEEDS_FIXES (truth: 1), ChatGPhish "as a duplicate" → "as not applicable" per Permiso primary. Iter 5 (Opus, cap): NEEDS_FIXES (truth: 1), "11 languages" quantifier unsourced; softened to "multiple languages". Final state: CLEAN_AFTER_REMEDIATION. verification_residual_count: 0.

Unmatched action items (migrated)

  • Audit Entra Agent ID app roles before deploying AI agents in M365: Establish a baseline of which agent identities hold AgentIdentityBlueprint.AddRemoveCreds.All; alert on credential additions to blueprints by any non-provisioning identity (AuditLogs: "Update application – Certificates and secrets management"). Review agent-identity API scopes as part of your AI-workload onboarding process. Reference: Red Canary.
  • Hunt for VS Code Remote Tunnel (code --tunnel) and Cloudflare Quick Tunnel (cloudflared.exe) on managed endpoints: Kimsuky and other actors are using these legitimate services to establish persistent C2 without fixable IP/domain blocks. Alert on VS Code CLI with --tunnel argument from non-developer endpoint profiles; alert on cloudflared.exe not in the approved software baseline. Reference: The Hacker News.
  • Update Marimo notebook to ≥ 0.23.0 and restrict internet-accessible notebook deployments: CVE-2026-39987 pre-auth RCE was exploited by an LLM-agent-driven actor to pivot to database exfiltration in under one hour. Also audit AWS CloudTrail for Secrets Manager GetSecretValue calls from notebook or unfamiliar IP space. Reference: Sysdig TRT.

Migrated from briefs/2026-05-30.md (v2).