7 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
01Kaspersky: Kimsuky's new PebbleDash backdoors hide C2 behind tunnels and take over hosts through VSCode Remote Tunnels. Kaspersky's Securelist report of 2026-05-14 analyzes Kimsuky (APT43, Ruby Sleet) campaigns with droppers detected from August 2025 to February 2026: spear-phishing attachments disguised as documents (JSE, SCR, PIF) install the PebbleDash-family backdoors HelloDoor (Rust, early-stage, C2 behind TryCloudflare), httpMalice, the MemLoad loader and httpTroy, and the AppleSeed and HappyDoor malware. For post-exploitation the group runs a legitimate VSCode CLI tunnel authorized with its own GitHub account or deploys DWAgent tied to its account; targets are mainly South Korean, and PebbleDash also hit defense organizations in Brazil and Germany. →
02DHTMLX PDF Export Module before 0.7.6 executes attacker JavaScript sent to it without authentication (CVE-2026-41553). CERT Polska disclosed three DHTMLX flaws on 2026-05-15: CVE-2026-41553 lets an unauthenticated attacker inject JavaScript into the data parameter of the PDF Export Module used by DHTMLX Gantt and Scheduler, which Node.js then executes (CVSS 4.0 10.0), while CVE-2026-41552 in the same module and CVE-2026-7182 in Diagram's export module let crafted HTML pull local server files into the generated PDF (9.2 each). The PDF Export Module from 0.3.3 is fixed in 0.7.6 and Diagram from 1.0.0 in 1.1.1; no source reports exploitation. →
03Funnel Builder for WooCommerce flaw let attackers plant fake Google Tag Manager card skimmers on every checkout page. Sansec reported on 2026-05-14 active exploitation of Funnel Builder by FunnelKit, a WooCommerce checkout plugin used on 40,000+ stores, whose public checkout endpoint before version 3.15.0.3 let unauthenticated requests write the plugin's External Scripts setting; attackers planted a fake Google Tag Manager loader that pulls a store-tailored card skimmer over WebSocket. The flaw is tracked as CVE-2026-47100 (CVSS 4.0 8.7); FunnelKit pushed the fix to older branches through WordPress.org automatic updates, saw signs of compromise on 3 of more than 1,300 sites it monitored, and said the exploitation path was closed. →
CERT Polska coordinated the disclosure of CVE-2026-44088 in SzafirHost from Krajowa Izba Rozliczeniowa (KIR) on 2026-05-15: SzafirHost verifies the signature of a downloaded JAR file with JarInputStream, which reads from the beginning of the file, but loads classes with JarFile/URLClassLoader, which reads the ZIP Central Directory from the end (CERT Polska, 2026-05-15). An attacker who combines a genuine signed JAR with a malicious ZIP file therefore passes verification while the malicious class is loaded, which CERT Polska says can lead to remote code execution; all versions before 1.2.1 are affected and version 1.2.1 fixes the flaw (CERT Polska, 2026-05-15). ENISA's EUVD carries CERT-PL's CVSS 4.0 base score of 8.6, with a network attack vector, no privileges and active user interaction (ENISA EUVD-2026-30512). SzafirHost is the application that, with KIR's browser plugin and Java, signs documents and tax declarations with a qualified signature; accounting software prompts users to install the plugin and SzafirHost at the first signing attempt (wFirma help centre). CERT Polska does not say how a malicious JAR would reach SzafirHost, and no source reports exploitation.
SzafirHost verifies the signature of the downloaded JAR file using class JarInputStream (reading from the beginning of the file), but loads classes using class JarFile/URLClassLoader (reading the Central Directory from the end). It can lead to remote code execution by allowing an attacker to combine a genuine, signed JAR file with a malicious ZIP file, causing the verification to pass but the malicious class to be loaded.
Links to install the KIR plugin and the Szafir Host application appear during the first signing attempt if they have not yet been installed on the device.
Where no such agreement exists, a qualified electronic signature based on a qualified certificate under foreign law is not recognised as being equivalent to a qualified electronic signature under Swiss law.
Switzerland does not recognise Polish qualified signatures under an eIDAS-equivalent framework: no mutual recognition agreement exists between Switzerland and the EU, and without one a foreign qualified signature is not equivalent to a qualified signature under Swiss law (BAKOM). CERT Polska states the impact as remote code execution from combining a genuine signed JAR with a malicious ZIP, in all SzafirHost versions before 1.2.1 (CERT Polska, 2026-05-15); it describes no attack on the download path, no silent signing of fraudulent documents and no list of public-sector workflows that depend on the helper. The CVSS 4.0 base score of 8.6 is CERT-PL's as the numbering authority, carried by ENISA's EUVD (ENISA EUVD-2026-30512).
threat17 May 05:00Zsingle-source · national CERTOpen finding →
Sansec reported on 2026-05-14 that it was tracking active attacks against Funnel Builder by FunnelKit, a checkout and upsell plugin used on 40,000+ WooCommerce stores, in which all versions before 3.15.0.3 let unauthenticated attackers inject arbitrary JavaScript into every checkout page (Sansec, 2026-05-14). The plugin's public checkout endpoint lets a request pick which internal method to run without checking the caller's permissions or restricting the methods, so an unauthenticated request can reach the method that writes the plugin's global settings, and whatever sits in the External Scripts setting is printed onto every Funnel Builder checkout page (Sansec, 2026-05-14). In at least one case Sansec found a payload posing as a Google Tag Manager loader beside the store's legitimate marketing tags: at page load it decodes a base64 string and loads that external script, which opens a WebSocket to the attacker's server and receives a skimmer tailored to the store that steals card numbers, CVVs, billing addresses and other personal details at checkout (Sansec, 2026-05-14). The flaw is tracked as CVE-2026-47100 and rated 8.7 in CVSS 4.0 (Sansec, 2026-05-14; ENISA EUVD-2026-30936).
FunnelKit shipped version 3.15.0.3 on 2026-05-14 (BleepingComputer, 2026-05-15), released the fix within 36 hours of the first report, pushed it to older release branches through WordPress.org so many stores updated without manual action, and DNS-blocked the attacker domains it knew (Sansec, 2026-05-14). In a statement added to The Hacker News report on 2026-05-31, FunnelKit said the exploitation path had been closed and that of more than 1,300 sites it monitored in the nine days after disclosure, three showed signs of compromise before its mitigations took effect (The Hacker News, 2026-05-16).
Triage: Sansec calls skimmers dressed as analytics tags a recurring Magecart pattern, because reviewers skip familiar tracking code; the planted loader names Google Analytics but at page load decodes a base64 string and loads its script from another host (Sansec, 2026-05-14).
All versions before 3.15.0.3 let unauthenticated attackers inject arbitrary JavaScript into every checkout page on the store.
Funnel Builder includes a public checkout endpoint that lets the request pick which internal method to run. Older releases never check the caller's permissions or restrict which methods are allowed.
Stores running version 3.15.0.3 or later, including those automatically updated through WordPress.org, are no longer vulnerable to this issue and are not exposed to the previously reported attack.
The overall impact appears to have been limited. During the nine days following disclosure, our monitoring covered a sample of more than 1,300 sites. Of those, only three showed signs of compromise before our mitigations took effect.
The flaw is tracked as CVE-2026-47100 and rated 8.7 in CVSS 4.0 (Sansec, 2026-05-14; ENISA EUVD-2026-30936). The 40,000+ figure is the number of stores using the plugin, not of exploited stores: Sansec found the fake Google Tag Manager loader in at least one case (Sansec, 2026-05-14), and FunnelKit said in a statement added to The Hacker News report on 2026-05-31 that of more than 1,300 sites it monitored in the nine days after disclosure, three showed signs of compromise before its mitigations took effect, and that the exploitation path had been closed after the fix was pushed to older branches through WordPress.org automatic updates (The Hacker News, 2026-05-16). The skimmer is tailored to each store and streamed over a WebSocket that the loader opens in the shopper's browser; no source describes WebSocket connections from the WordPress PHP process or exposure of Swiss or EU public portals, and the check FunnelKit advises is a review of the plugin's External Scripts setting (Sansec, 2026-05-14).
CERT Polska coordinated the disclosure of three vulnerabilities in DHTMLX software on 2026-05-15, reported by Łukasz Jaworski and Tomasz Holeksa of Pentest Limited (CERT Polska, 2026-05-15). In CVE-2026-41553 the PDF Export Module used by DHTMLX Gantt and Scheduler does not sanitize its data parameter, so an unauthenticated attacker can inject JavaScript that Node.js processes and executes, which can lead to server compromise; CVE-2026-41552 in the same module and CVE-2026-7182 in Diagram's export module (through the src attribute) lack HTML sanitization, so a crafted payload can include local files from the server in the generated PDF (CERT Polska, 2026-05-15). ENISA's EUVD carries CERT-PL's CVSS 4.0 scores of 10.0 for CVE-2026-41553 and 9.2 for each file-read flaw, all with a network vector, no privileges and no user interaction (ENISA EUVD-2026-30537; ENISA EUVD-2026-30538; ENISA EUVD-2026-30539). PDF Export Module version 0.7.6 fixes CVE-2026-41552 and CVE-2026-41553 and Diagram version 1.1.1 fixes CVE-2026-7182 (CERT Polska, 2026-05-15), and EUVD lists the module from 0.3.3 and Diagram from 1.0.0 as affected below those releases (ENISA EUVD-2026-30537; ENISA EUVD-2026-30539); DHTMLX's changelog for release 0.7.6 records fixed remote code execution and file read vulnerabilities (DHTMLX PDF Export Module changelog). No source reports exploitation.
An unauthenticated attacker can inject the malicious JavaScript code to the parameter whose value is processed by Node.js and subsequently executed.
Vulnerabilities CVE-2026-41552 and CVE-2026-41553 were fixed in PDF Export Module version 0.7.6.
CERT Polska's advisory names the vendor, the products and the affected ranges and says nothing about where DHTMLX components are deployed, so the claim that they are widely embedded in EU e-government, healthcare scheduling and municipal planning tools has no source (CERT Polska, 2026-05-15). The disclosure covers the PDF Export Module used by DHTMLX Gantt and Scheduler and the export module of DHTMLX Diagram, not the scheduling and diagram libraries as a whole (CERT Polska, 2026-05-15). CVE-2026-41552 and CVE-2026-7182 are path-traversal flaws that let crafted HTML include local server files in the generated PDF rather than code execution, with CVSS 4.0 scores of 9.2 (ENISA EUVD-2026-30538; ENISA EUVD-2026-30539). The PDF Export Module is affected from release 0.3.3 and fixed in 0.7.6, and Diagram from 1.0.0 and fixed in 1.1.1 (ENISA EUVD-2026-30537; CERT Polska, 2026-05-15).
On Day Two of Pwn2Own Berlin 2026, "Orange Tsai (@orange_8361) of DEVCORE Research Team chained 3 bugs to achieve Remote Code Execution as SYSTEM on Microsoft Exchange, earning $200,000 and 20 Master of Pwn points" (Zero Day Initiative, 2026-05-15). BleepingComputer reported the same result (BleepingComputer, 2026-05-15). The bugs were zero-days disclosed to Microsoft at the contest, which gives vendors 90 days to patch (BleepingComputer, 2026-05-15); neither source says whether the chain needs authentication. CVE-2026-42897, the exploited Outlook Web Access cross-site scripting flaw, had already been disclosed by Microsoft (Microsoft Security Response Center), so it is not one of the contest's zero-days.
Contradiction: for CVE-2026-62911, Microsoft describes an authorized attacker (CVSS 8.0) (MSRC CVE-2026-62911), while Franky's Web says the attacker does not need valid credentials (Franky's Web, 2026-08-27) and MB VRED titles its analysis a "Pre-Auth RCE" (MB VRED, 2026-08-13).
Orange Tsai (@orange_8361) of DEVCORE Research Team chained 3 bugs to achieve Remote Code Execution as SYSTEM on Microsoft Exchange, earning $200,000 and 20 Master of Pwn points.
He demonstrated the vulnerability at Pwn2Own Berlin 2026, where it was part of a chain of three vulnerabilities that together allowed remote code execution with SYSTEM privileges on Exchange.
ZDI describes the result as remote code execution as SYSTEM on Microsoft Exchange through three chained bugs (Zero Day Initiative, 2026-05-15); neither ZDI nor BleepingComputer says the chain works without authentication, and the earlier quotation was not ZDI's wording. The chain is not CVE-2026-42897, which was a separate, exploited cross-site scripting flaw in Outlook Web Access (Microsoft Security Response Center). The priority moves from notable to routine: the contest result itself leaves nothing to apply beyond keeping Exchange current as Microsoft's fixes ship. CVE-2026-62911 was fixed in the August 2026 Exchange security update (MSRC CVE-2026-62911), and Franky's Web reports that Orange Tsai demonstrated it at Pwn2Own Berlin 2026 as part of this chain, and that working exploit code surfaced in late August (Franky's Web, 2026-08-27).
Microsoft's record for CVE-2026-62911 lists the August 2026 fixed builds, credits "Orange Tsai (@orange_8361) of DEVCORE Research Team working with TrendAI Zero Day Initiative" and describes "an authorized attacker", but does not mention Pwn2Own (MSRC CVE-2026-62911). The identification of CVE-2026-62911 as one of the contest chain's three bugs comes from Franky's Web (Franky's Web, 2026-08-27) and a public analysis of the demo (MB VRED, 2026-08-13); neither ZDI nor Microsoft states it. Franky's Web says the attacker does not need valid credentials, which disagrees with Microsoft's description of an authorized attacker (Franky's Web, 2026-08-27).
F5 announced on 2026-05-13 the fixes in its quarterly security notification (F5 K000160932; SecurityWeek, 2026-05-14), among them CVE-2026-41225 in BIG-IP iControl REST: an authenticated attacker with at least the Manager role can create configuration objects that allow running arbitrary commands (ENISA EUVD-2026-29986). F5 says the flaw lets a highly privileged attacker with network access to the iControl REST endpoint through the BIG-IP management port or self IP addresses escalate privileges or bypass Appliance mode restrictions, crossing a security boundary in Appliance mode deployments, and that there is no data-plane exposure (SecurityWeek, 2026-05-14). F5 scores it CVSS 4.0 8.6 (ENISA EUVD-2026-29986; SecurityWeek, 2026-05-14), and NCSC-NL lists a CVSS 3.1 score of 9.1 with a changed scope (NCSC-NL NCSC-2026-0162, 2026-05-15). Affected are BIG-IP 21.0.0 before version 21.0.0.2, 17.5.0 before version 17.5.1.6 and 17.1.0 before version 17.1.3.2, with the 16.1 branch listed as affected and, being past end of technical support, not evaluated (ENISA EUVD-2026-29986). SecurityWeek counts over 19 high-severity and 32 medium-severity fixes across BIG-IP, BIG-IQ and NGINX in the same notification, none of them exploited in the wild as of 2026-05-14 (SecurityWeek, 2026-05-14), and NCSC-NL's advisory lists 43 CVEs for BIG-IP and BIG-IQ (NCSC-NL NCSC-2026-0162, 2026-05-15).
A vulnerability exists in iControl REST where a highly privileged, authenticated attacker with at least the Manager role can create configuration objects that allow running arbitrary commands.
This vulnerability may allow a highly privileged attacker with network access to the affected iControl REST endpoint through the BIG-IP management port or self IP addresses to escalate their privileges or bypass Appliance mode restrictions.
None of these vulnerabilities appears to have been exploited in the wild.
CVE-2026-41225 is listed for BIG-IP only, in 21.0.0 before version 21.0.0.2, 17.5.0 before version 17.5.1.6 and 17.1.0 before version 17.1.3.2, with the 16.1 branch past end of technical support and not evaluated (ENISA EUVD-2026-29986); F5 published it on 2026-05-13 (SecurityWeek, 2026-05-14). The prerequisite is an authenticated account with at least the Manager role and network access to iControl REST through the management port or a self IP address, and F5 says there is no data-plane exposure (SecurityWeek, 2026-05-14); no source describes a credential-theft chain, a European public-sector deployment base or monitoring guidance. The secondary list of CVE ids grouped by flaw class did not match NCSC-NL's advisory, which gives the cleartext SSH-password exposure and the SNMP-object privilege escalation as separate flaws with their own ids (NCSC-NL NCSC-2026-0162, 2026-05-15). F5's own score is CVSS 4.0 8.6 (ENISA EUVD-2026-29986); the 9.1 comes from NCSC-NL as a CVSS 3.1 score with a changed scope (NCSC-NL NCSC-2026-0162, 2026-05-15).
Kaspersky's report covers Kimsuky (also APT43, Ruby Sleet) campaigns whose droppers it detected between August 2025 and February 2026: spear-phishing emails deliver compressed attachments disguised as documents, with lure names carrying a document extension before .jse or .pif (Kaspersky Securelist, 2026-05-14). JSE droppers decode at least two Base64 blobs with JScript into a location such as C:\ProgramData, open the lure, then decode the payload a second time with certutil from a hidden PowerShell window and run it through regsvr32 or rundll32, while the SCR and PIF droppers decrypt theirs with XOR (Kaspersky Securelist, 2026-05-14). HelloDoor, a Rust DLL backdoor first seen in August 2025 that Kaspersky judges early-stage, persists through an HKCU Run value that launches regsvr32 and talks to a C2 hosted through TryCloudflare, which makes the infrastructure behind it hard to trace; httpMalice (version 1.9 over HTTP or HTTPS, version 1.8 over the Dropbox API) persists as a service when elevated and through an HKCU Run value otherwise, profiles the host, captures the screen, uploads archived directories and encrypts its traffic with ChaCha20 (Kaspersky Securelist, 2026-05-14). MemLoad runs anti-VM checks and reconnaissance, persists through scheduled tasks that run regsvr32 on its DLL every minute, and reflectively loads the RC4-encrypted httpTroy backdoor in memory for long-term access and exfiltration; the AppleSeed Spy variant collects documents, screenshots, keystrokes and the GPKI certificate directory (Kaspersky Securelist, 2026-05-14).
For post-exploitation the group downloads the legitimate VSCode CLI, unpacks it into C:\Users\Public and runs code.exe tunnel, completing the GitHub device-code authorization with its own account so the host then talks to Microsoft-owned servers, or installs DWAgent from C:\ProgramData with a configuration linked to its own account; it also tunnels through Ngrok, and some samples are signed with certificates stolen from South Korean organizations (Kaspersky Securelist, 2026-05-14). Targets are mainly South Korean public and private entities, and PebbleDash victims also include Brazilian defense organizations and a German defense firm (Kaspersky Securelist, 2026-05-14).
Triage: VSCode Remote Tunnels and DWAgent are legitimate remote-access tools; in Kimsuky's use the VSCode CLI is unpacked into C:\Users\Public and started by a script dropper or installer rather than by the user, and DWAgent arrives with a preset configuration from a dropper that unpacks it into C:\ProgramData (Kaspersky Securelist, 2026-05-14).
Our monitoring indicates various strategic updates to the group’s arsenal, including the use of VSCode Tunneling, Cloudflare Quick Tunnels, DWAgent, large language models (LLMs), and the Rust programming language.
Kimsuky obtains initial access to target systems by delivering spear-phishing emails containing malicious attachments disguised as documents.
Since it has limited capabilities and a simplistic communication mechanism, the backdoor is most probably in the early stages of development.
Kimsuky mainly targets South Korean entities. However, PebbleDash attacks were also seen in Brazil and Germany.
httpMalice is not an HTTP-only loader but a PebbleDash-based backdoor that talks over HTTP or HTTPS in version 1.9 and through the Dropbox API in version 1.8, MemLoad is the loader that runs anti-VM checks and reconnaissance before reflectively loading httpTroy, and httpTroy is the backdoor for long-term access and data exfiltration (Kaspersky Securelist, 2026-05-14). The report gives observations rather than detection guidance, so detection follows the dropper, persistence and tunnel artifacts it documents (Kaspersky Securelist, 2026-05-14). HelloDoor is judged early-stage, and its TryCloudflare-hosted C2 makes its infrastructure difficult to trace; Kimsuky also tunnels through VSCode Tunneling and Ngrok (Kaspersky Securelist, 2026-05-14). Targets are mainly South Korean, AppleSeed most often government bodies and PebbleDash the defense, military and medical sectors, and PebbleDash victims also include Brazilian defense organizations and a German defense firm over several years (Kaspersky Securelist, 2026-05-14).
Pwn2Own Berlin 2026, ZDI's second Berlin contest after the inaugural 2025 event, ran at OffensiveCon from 14 to 16 May (Zero Day Initiative, 2026-03-12) (BleepingComputer, 2026-05-15). For 2026 ZDI split its AI category into AI Databases, Coding Agents, Local Inferences and NVIDIA (Zero Day Initiative, 2026-03-12). Targets ran fully patched in their default configurations (Zero Day Initiative, contest rules), and vendors have 90 days to patch after the zero-days are disclosed at the contest (BleepingComputer, 2026-05-15). Before it began, Hackread wrote that it had reportedly run out of space and that some researchers turned away were releasing their findings instead (Hackread, 2026-05-12).
The final totals were $1,298,250 for 47 unique zero-days; DEVCORE won Master of Pwn with 50.5 points and $505,000, ahead of STARLabs SG (25 points, $242,500) and Out Of Bounds (12.75 points, $95,750) (Zero Day Initiative, 2026-05-16).
Enterprise and platform targets. Orange Tsai of DEVCORE chained four logic bugs for a Microsoft Edge sandbox escape ($175,000) on Day One (Zero Day Initiative, 2026-05-14) and three bugs for remote code execution as SYSTEM on Microsoft Exchange ($200,000) on Day Two (Zero Day Initiative, 2026-05-15); neither ZDI nor BleepingComputer states whether the chain needs authentication. splitline of DEVCORE chained two bugs against Microsoft SharePoint ($100,000), and Nguyen Hoang Thach of STARLabs SG used a memory corruption bug against VMware ESXi with the cross-tenant code execution add-on ($200,000) (Zero Day Initiative, 2026-05-16), an entry that runs code from a guest on the hypervisor (Zero Day Initiative, contest rules). Windows 11 and Red Hat Enterprise Linux for Workstations fell to privilege escalations (improper access control, heap overflow, use-after-free, integer overflow and race-condition bugs), and NV Container Toolkit to chompie ($50,000) and to 0xDACA and Noam Trobishi with a use-after-free ($25,000) (Zero Day Initiative, 2026-05-14) (Zero Day Initiative, 2026-05-15) (Zero Day Initiative, 2026-05-16).
This year, we’re expanding this and splitting it into multiple different categories: AI Databases, Coding Agents, Local Inferences, and a separate category for NVIDIA products.
ZDI and BleepingComputer report the Exchange result as three chained bugs giving remote code execution as SYSTEM, and neither says the chain works without authentication (Zero Day Initiative, 2026-05-15) (BleepingComputer, 2026-05-15). ZDI ran no "AI Agents" category: for 2026 it split its existing AI category into AI Databases, Coding Agents, Local Inferences and NVIDIA (Zero Day Initiative, 2026-03-12), and not every AI target fell: an Oracle Autonomous AI Database attempt failed, while Chroma was a two-bug success rather than a collision (Zero Day Initiative, 2026-05-14). The pages report results only; they carry no patch forecast, no attack-pattern analysis of the AI targets and no hardening advice. The release of findings by researchers turned away from the contest was reported by Hackread before the contest as the researchers' own action, not ZDI's (Hackread, 2026-05-12). Vendors have 90 days to patch after the zero-days are disclosed at the contest (BleepingComputer, 2026-05-15), and the NV Container Toolkit researcher is Noam Trobishi (Zero Day Initiative, 2026-05-15).
2026-05-17-4381863a· Claude Opus 4.7 · window 40 h · 7 entries published
Items dropped:
GitLab CE/EE 18.11.3 security release, 25 CVEs including CVE-2026-7481 / CVE-2026-7377 / CVE-2026-6073 (CVSS 8.7 stored XSS). Dropped from § 2 because no CVE in the cluster clears any § 2 inclusion gate: not CISA-KEV-listed, no ENISA-EUVD exploited=true entry, top CVSS 8.7 (< 9.0), no public PoC, no in-the-wild exploitation reported. The XSS cluster is post-auth (Developer role required) with victim interaction prerequisite. Recommend operators apply through normal GitLab patch cadence; flagging would be over-weighted given today's competing operational signal. Source for reference: NCSC-NL NCSC-2026-0161, 2026-05-15.
FamousSparrow (UAT-9244) Azerbaijani oil & gas three-wave Exchange intrusion (Bitdefender, 2026-05-13). Surfaced by S3 and S4 but the primary publication is from 2026-05-13, outside both the 40 h recency window and the 72 h developing window. The 2026-05-14 daily deep dive already covered FamousSparrow against the Azerbaijani energy operator; no material new development this run. Source for reference: Bitdefender Business Insights, 2026-05-13.
Single-source items: § 3 Kimsuky / PebbleDash / HelloDoor (Kaspersky Securelist as sole source), included under the HIGH-reliability research-lab carve-out; explicitly flagged [SINGLE-SOURCE] in the item.
Items included with reduced confidence (window edge): § 3 Kimsuky item is at the edge of the 72 h developing window (Securelist publication 2026-05-14, ~62 h before run start), included because the technical novelty (Rust-based PebbleDash variant + TryCloudflare quick-tunnel C2) is defender-actionable and not previously covered.
Contradictions: none surfaced this run.
Sub-agent timing: all four cti-research sub-agents (S1, S2, S3, S4) returned within budget. S1 → S4 durations 489 / 470 / 678 / 516 seconds, all well under the 30-min hard cap. S3 wrote S3.ended_at but its findings.S3.yaml was finalised at 04:19:21 UTC, with the assistant-text return delivered ~2 minutes after the disk artefact landed (no operational impact, main agent waits on the disk artefact, not the assistant-text return).
Coverage gaps: databreaches-net (5 consecutive runs failing 403, Cloudflare anti-bot, no Wayback corroboration available; covered via primary regulator notices and victim disclosures where present); inside-it-ch (5 runs 403, Cloudflare-protected, bridge fetches blocked); cisco-psirt RSS endpoint returned 404 in this run (last known good 2026-05-15; main vendor PSIRT page reachable, advisory list not); cert-fr (no new items since 2026-05-13 in the AVI/ACT feeds); cert-eu (most recent advisory dated 2026-05-13, outside the 40 h window); talos (bridge known-403, not attempted in this run, pivoted to other channels; pre-window content only); sophos-xops (HTTP 503 on blog fetch in this run); dfirreport (EtherRat + TukTuk flash alert published 2026-05-11, outside 72 h developing window). One candidate source surfaced by S3, sansec-research (Sansec, sansec.io, Magecart and e-commerce skimming primary research), promoted to status: "candidate" in sources/sources.json per the one-candidate-per-run rule.
Verification: five Phase 5.7 iterations ran (Opus / Sonnet-alt / Opus / Sonnet-alt / Opus rotation per v2.47); the brief publishes at the v2.46 cap-breach safety valve with the final iteration NEEDS_FIXES (truth=4, editorial=0). Iter1 found 5 truth + 1 editorial + 1 advisory (Pwn2Own attribution drift, F5 source-order, F5 quantifier); iter2 found 3 truth + 1 advisory (Satoki Tsuji attribution, CVSS dual-scale confusion 8.6 vs 9.1, OtterSec LM Studio technique over-claim); iter3 found 4 truth + 2 editorial + 1 advisory (SzafirHost overreach on Polish public-sector specifics, Swiss procurement portal claim, "dominant" quantifier, SecurityWeek phrasing, NCSC-NL CSAF source-attribution gap, Day 1 Orange Tsai Edge omission); iter4 found 1 truth + 1 editorial (NCSC-NL SPA URLs swapped for canonical CSAF JSON URLs, CWE-250 → CWE-648 per NVD). Iter5 residuals retained as the cap-breach record: F1 quote re-attributed (Sansec → The Hacker News citing Sansec, applied); F4 DHTMLX src qualifier scoped to CVE-2026-7182 only (applied). Iter5 findings F2 and F3 were verifier false-positives, main agent re-verified the NCSC-NL CSAF JSON directly (tools/fetch_source.py ncsc-nl csaf NCSC-2026-0162) showing 43 unique CVEs in the BIG-IP / BIG-IQ scope (not 23 as iter5 claimed) and all of CVE-2026-42930, CVE-2026-42924, CVE-2026-42406, CVE-2026-41953 present in the CSAF (iter5 claimed CVE-2026-42406 and CVE-2026-41953 absent). The brief preserves the 43-count and the four-CVE injection cluster as written. verification_residual_count = 4 records iter5's raw counts before this rebuttal-by-direct-CSAF-re-verification.