01A forged vHub peer can plant an SSH key on Cisco SD-WAN controllers without credentials; Talos ties exploitation to UAT-8616. CVE-2026-20182 (CVSS 10.0) lets an unauthenticated attacker bypass peering authentication on Cisco Catalyst SD-WAN Controller, Manager and Validator and plant an SSH key for NETCONF access. Cisco saw limited exploitation in May 2026, which Talos attributes to UAT-8616; CISA added it to its KEV catalog on 2026-05-14 under Emergency Directive 26-03. There is no workaround, and a compromised system needs TAC remediation beyond the upgrade. →
02CVE-2026-46300 (Fragnesia): CISA lists the Linux kernel root flaw as exploited, chained with PHP2Shell. CVE-2026-46300 ("Fragnesia", CVSS 7.8) is a local privilege escalation in the Linux kernel's xfrm ESP-in-TCP path, one of three CVEs Red Hat groups as "Dirty Frag"; CISA added it to its Known Exploited Vulnerabilities catalog on 2026-10-28 and says it can be chained with the PHP2Shell flaws CVE-2026-109618 and CVE-2026-109619 to implant PHP webshells and run commands as root, and Dream Research reports such a chain against 25 Palo Alto Networks PAN-OS firewall and management servers and a US government-affiliated network. Its public exploit overwrites /usr/bin/su in the page cache, and Red Hat confirms RHEL kernels are affected (CISA KEV, 2026-10-28 · Dream Research, 2026-10-29 · Red Hat RHSB-2026-003). →
03An 18-year-old NGINX rewrite-module heap overflow has a public PoC, and VulnCheck honeypots saw exploitation attempts. NGINX "NGINX Rift" CVE-2026-42945 (CVSS 9.2/4.0): 18-year-old heap overflow in ngx_http_rewrite_module now has a public PoC and VulnCheck saw exploitation attempts against honeypots; NCSC-CH advisory published 2026-05-15; affects NGINX 0.6.27–1.30.0, Plus R32–R36, Kubernetes Ingress Controller, and multiple F5 products (versions per depthfirst; NCSC-CH Security Hub #12575, 2026-05-15). →
ESET attributes new activity since March 2026 to FrostyNeighbor, also known as Ghostwriter, UNC1151, UAC-0057, TA445, PUSHCHA or Storm-0257, a long-running cyberespionage group apparently aligned with the interests of Belarus, using links in spearphishing PDFs to target governmental organisations in Ukraine (ESET, 2026-05-14). In ESET's telemetry the group's Ukrainian targeting focuses on military, defence and government, while its victimology in Poland and Lithuania is broader, spanning industry and manufacturing, healthcare and pharmaceuticals, logistics and many governmental organisations (ESET, 2026-05-14).
The lure is a blurred PDF impersonating Ukrtelecom whose download button leads to the group's delivery server; visitors outside Ukraine receive a benign PDF of the same name, while Ukrainian IP addresses receive a RAR archive holding a JavaScript first stage that shows a decoy, re-runs itself, drops a base64-embedded JavaScript PicassoLoader and builds a scheduled task from an XML template the C2 serves as a fake JPG (ESET, 2026-05-14). PicassoLoader collects the username, computer name, OS version, boot time, current time and running processes, posts them over HTTPS every 10 minutes and evaluates any response longer than 100 bytes (ESET, 2026-05-14). Operators very likely decide by hand whether the victim is of interest; if so, the C2 returns a third-stage dropper that copies rundll32.exe into %ProgramData% under another name, writes the base64-decoded Cobalt Strike beacon DLL beside it, and registers in the HKCU Run key a LNK that starts the copied rundll32 with the DLL's export (ESET, 2026-05-14).
Triage:rundll32.exe is a normal Windows binary when it runs from the system directory; a copy of it under a different name in %ProgramData%, which ESET says is very likely meant to bypass detection rules, is the signal (ESET, 2026-05-14).
Since March 2026, we have detected new activities that we attributed to FrostyNeighbor, using links in malicious PDFs sent via spearphishing attachments to target governmental organizations in Ukraine.
The decision whether or not to deliver a payload is very likely manually performed by the operators, based on the collected information to decide if the victim is of interest.
The payload is only delivered after server-side victim validation, combining automated checks of the requesting user agent and IP address with the manual validation by the operators.
The activity ESET documents since March 2026 targets governmental organisations in Ukraine (ESET, 2026-05-14); the industrial, healthcare and pharmaceutical, logistics and governmental victims in Poland and Lithuania come from ESET's introduction to the group's broader victimology, not from this campaign wave. The chain also persists: the first stage builds a scheduled task from a downloaded XML template to run PicassoLoader, and the Cobalt Strike stage copies rundll32.exe into %ProgramData% under another name and starts it from a LNK registered in the HKCU Run key (ESET, 2026-05-14), which gives defenders host artefacts to hunt where geofencing hides the network side.
On 2026-05-12 a GitHub repository appeared with what Datadog Security Labs assesses as the complete source code of the Shai-Hulud offensive framework attributed to TeamPCP; its README reads "Love - TeamPCP" and "Change keys and C2 as needed", its commits are authored as TeamPCP_OSS, and GitHub removed it shortly after publication, though not before it was forked several times (Datadog Security Labs, 2026-05-13). Datadog maps the source directly to compiled payloads seen in the wild, including those of the TanStack and UiPath compromises, and finds 19 of the 22 TeamPCP techniques described in earlier reporting implemented in it (Datadog Security Labs, 2026-05-13).
The modular TypeScript and Bun toolkit exits on systems with a Russian locale, then reads OS-specific lists of more than 100 sensitive file paths, runs gh auth token and captures the process environment, and in GitHub Actions on Linux feeds a script to sudo python3 that reads the Runner.Worker process memory for values marked secret; further providers walk the AWS credential chain including instance and container metadata, read every Kubernetes secret in every namespace and enumerate Vault (Datadog Security Labs, 2026-05-13). Data leaves AES-256-GCM encrypted over HTTPS to a campaign domain, with a fallback domain found through signed GitHub commits and, failing that, new public GitHub repositories used as dead-drops (Datadog Security Labs, 2026-05-13). Stolen GitHub tokens are used to commit VS Code folder-open tasks and Claude Code SessionStart hooks to every branch of the victim's repositories, stolen npm tokens to republish the victim's packages with a preinstall loader, and, in a release workflow it was hardcoded to target, the workflow's OIDC token to publish a poisoned package with genuine Sigstore provenance; where it attaches a stolen token it installs a LaunchAgent or systemd user service that checks the token every minute and runs rm -rf ~/ once it is revoked (Datadog Security Labs, 2026-05-13).
The repository's README explicitly states: "Love - TeamPCP" and "Change keys and C2 as needed."
This appears to have been removed by GitHub shortly after publishing, though not in time to prevent multiple forks.
the framework installs a daemon (gh-token-monitor) that runs rm -rf ~/ if your token is revoked. Disable it first, then rotate credentials.
The published repository's README reads "Love - TeamPCP" and "Change keys and C2 as needed", its commits are authored as TeamPCP_OSS, and GitHub removed the repository shortly after publication, though not before it was forked, so the code remains available to other actors (Datadog Security Labs, 2026-05-13). Datadog's analysis describes no pull-request exploitation, pull_request_target targeting or expression-injection capability in the code. The framework harvests credentials where it already runs, reads Runner.Worker memory, poisons repository branches and npm packages with stolen tokens, and uses a workflow's own OIDC token only inside a release workflow and for a package it was hardcoded to target, so that path was manually targeted while the token-driven spread is opportunistic (Datadog Security Labs, 2026-05-13).
The analysis also documents what defenders need for a compromise check: an exit on Russian-locale systems, a daemon that deletes the home directory when a stolen GitHub token is revoked and must be disabled before rotation, and the hook files, commits and npm releases it leaves behind (Datadog Security Labs, 2026-05-13).
On 2026-05-13 BleepingComputer and The Register reported that the researcher Nightmare Eclipse (also Chaotic Eclipse) had published exploit code for two Windows zero-days, YellowKey and GreenPlasma (BleepingComputer, 2026-05-13) (The Register, 2026-05-13). The researcher's earlier BlueHammer (CVE-2026-33825) and RedSun drops began to be exploited in the wild shortly after release (BleepingComputer, 2026-05-13).
YellowKey is a BitLocker bypass that needs physical access. The researcher says it affects Windows 11 and Windows Server 2022/2025 and works by placing crafted FsTx files on a USB drive or the EFI partition, rebooting into the Windows Recovery Environment (WinRE) and holding CTRL to trigger a shell (BleepingComputer, 2026-05-13). Will Dormann reproduced it with the files on a USB drive but not on the EFI partition: to boot WinRE, Windows replays the NTFS logs in \System Volume Information\FsTx directories on attached drives, the replay deletes X:\Windows\System32\winpeshl.ini, and WinRE then opens CMD.EXE instead of the recovery environment, with the disk still unlocked (BleepingComputer, 2026-05-13). Because it abuses TPM-only automatic unlock, the public exploit works only on the original device, not on a removed drive, and does not work with TPM+PIN (BleepingComputer, 2026-05-13). Microsoft tracks it as CVE-2026-45585, a security feature bypass by physical attack (CWE-77, CVSS 6.8), publicly disclosed and not exploited; its fixed builds are Windows 11 24H2 build 10.0.26100.8655, 25H2 build 10.0.26200.8655, 26H1 build 10.0.28000.2269 and Windows Server 2025 build 10.0.26100.32995, and its product table does not list Windows Server 2022 (MSRC CVE-2026-45585).
Contradiction: the researcher says an unpublished variant also defeats TPM+PIN (BleepingComputer, 2026-05-13); Microsoft states that "if you are using TPM+PIN the vulnerability is not exploitable" (MSRC CVE-2026-45585).
GreenPlasma is a local privilege escalation in CTFMON: an unprivileged user can create arbitrary memory-section objects within directory objects writable by SYSTEM, and the leaked PoC lacks the component needed for a full SYSTEM shell (BleepingComputer, 2026-05-13); in its current state it triggers a UAC consent prompt in default Windows configurations (The Register, 2026-05-13). Microsoft tracks it as CVE-2026-45586, a link-following flaw (CWE-59, CVSS 7.8) in the Windows Collaborative Translation Framework that lets an authorized local attacker gain SYSTEM privileges, released on 2026-06-09 with fixes for Windows 10, Windows 11 and Windows Server 2012 to 2025, publicly disclosed and not exploited (MSRC CVE-2026-45586).
Windows looks for \System Volume Information\FsTx directories on attached drives, and will replay any NTFS logs.
that the YellowKey exploit worked with the FsTx files on a USB drive but could not reproduce the bug using the EFI partition
Microsoft formally assigned CVE-2026-45585 to the BitLocker / WinRE bypass disclosed by "Nightmare Eclipse" on 2026-05-12 and confirmed there is still no security update. The MSRC update guide entry, published 2026-05-19, classifies it as CWE-77 (command injection in BitLocker / Windows Recovery Environment), CVSS 6.8 (AV:P/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H), with exploit-code maturity rated E:P (proof-of-concept) and remediation level RL:W (workaround only).
A pre-boot PIN avoids depending on the WinRE edit: the public exploit relies on TPM-only automatic unlock and does not work with TPM+PIN (BleepingComputer, 2026-05-13), and NCSC-CH advises that BIOS passwords and a BitLocker PIN mitigate the current public PoC (NCSC-CH Security Hub #12574, 2026-05-14).
Microsoft fixed both flaws in the June 2026 Windows updates of 2026-06-09. Revision 2 of the CVE-2026-45585 record "Added links to June 2026 Windows security updates", with fixed builds Windows 11 24H2 build 10.0.26100.8655, 25H2 build 10.0.26200.8655, 26H1 build 10.0.28000.2269 and Windows Server 2025 build 10.0.26100.32995 (MSRC CVE-2026-45585). GreenPlasma received its own identifier, CVE-2026-45586, released the same day with fixes for Windows 10, Windows 11 and Windows Server 2012 to 2025 (MSRC CVE-2026-45586), and BleepingComputer reported that the June 2026 Patch Tuesday fixed GreenPlasma, MiniPlasma and YellowKey (BleepingComputer, 2026-06-10). Before the fix, on 2026-05-21, Microsoft had replaced its manual WinRE steps with a script that removes autofstx.exe from the WinRE BootExecute value and re-seals WinRE so BitLocker trust remains intact (MSRC CVE-2026-45585). Microsoft marks neither flaw as exploited (MSRC CVE-2026-45585) (MSRC CVE-2026-45586).
Nextcloud's advisory for CVE-2026-45691 (Moderate, 5.9) says a session cookie created after successful password authentication but before TOTP completion could be reused as a Bearer token against DAV endpoints, granting read and write access while bypassing mandatory two-factor authentication (Nextcloud, 2026-05-13). The attacker therefore needs the user's password, which the advisory scores as low privileges with high attack complexity (Nextcloud, 2026-05-13). Server is affected from 32.0.0 and 33.0.0 and fixed in 32.0.9 and 33.0.3; Enterprise Server is affected from 29.0.0 through the 33 branch and fixed in 33.0.3, 32.0.9, version 31.0.14.5, version 30.0.17.9 and version 29.0.16.16 (Nextcloud, 2026-05-13). The same releases fix CVE-2026-45690, a second 2FA bypass by replaying a pending session token (Moderate, 5.9) (Nextcloud, 2026-05-15). The advisory lists no workaround, and CISA's Known Exploited Vulnerabilities catalogue did not list either flaw as of 2026-10-21 (CISA KEV catalog, 2026-10-21). BSI's WID-SEC-2026-1517 covers CVE-2026-45543, CVE-2026-45544, CVE-2026-45545 and CVE-2026-45691 from the batch (BSI, 2026-05-13).
Nextcloud's other advisories of that week concern separate apps: an SQL injection in the column-type parameter of the Tables app (CVE-2026-45545, High 8.2) (Nextcloud, 2026-05-13), a missing JWT signature check in ID4me handling of the User OIDC app (CVE-2026-45156, High 8.1) (Nextcloud, 2026-05-12), and user identifiers leaked through the Calendar app's attendee suggestion endpoint (CVE-2026-45286, Moderate 4.3) (Nextcloud, 2026-05-12).
A pre-2FA session cookie (created after successful password authentication but before TOTP completion) could be reused as a Bearer token to authenticate against DAV endpoints, granting read/write access and bypassing mandatory two-factor authentication.
It is recommended that the Nextcloud Enterprise Server is upgraded to 33.0.3, 32.0.9, 31.0.14.5, 30.0.17.9 or 29.0.16.16
The flaw is not exploitable before authentication: the reusable session cookie is created after successful password authentication, so an attacker needs the user's password (Nextcloud, 2026-05-13). Server is affected from 32.0.0 before 32.0.9 and from 33.0.0 before 33.0.3 (Nextcloud, 2026-05-13). CVE-2026-45690 is a second 2FA bypass fixed in the same releases (Moderate 5.9) (Nextcloud, 2026-05-15), not an SQL injection; the column-type SQL injection is CVE-2026-45545 in the Tables app, rated High 8.2 (Nextcloud, 2026-05-13), the User OIDC JWT signature bypass is rated High 8.1 (Nextcloud, 2026-05-12), and the Calendar attendee-suggestion leak is Moderate 4.3 (Nextcloud, 2026-05-12).
No cited source describes how widely Nextcloud is deployed in EU government or education, or states a public proof of concept or exploitation; with the password prerequisite, a Moderate score and no reported exploitation, the fix belongs in the regular patch cycle.
CVE-2026-20182 (CVSS 10.0, CWE-287) is an authentication bypass in the peering authentication of Cisco Catalyst SD-WAN Controller, Manager and, since Cisco's advisory revision of 2026-06-16, Validator, affecting all deployment types regardless of configuration (Cisco PSIRT). Rapid7 showed the mechanism on the Controller: the vdaemon DTLS control-plane service on UDP/12346 accepts a CHALLENGE_ACK claiming device type 2 (vHub) from a self-signed peer without verifying it, after which message type 14 (MSG_VMANAGE_TO_PEER) appends an attacker SSH key to /home/vmanage-admin/.ssh/authorized_keys, giving NETCONF access on TCP/830 as vmanage-admin (Rapid7, 2026-05-14). Rapid7 demonstrates the bypass with its own Metasploit module, submitted to the public Metasploit Framework repository (Rapid7, 2026-05-14). Cisco became aware of limited exploitation in May 2026 (Cisco PSIRT), and Talos attributes it to UAT-8616, a highly sophisticated actor that attempted to add SSH keys, modify NETCONF configurations and escalate to root (Cisco Talos, 2026-05-14). Talos separately tracks ten other clusters, distinct from UAT-8616, that exploit the earlier SD-WAN Manager flaws CVE-2026-20133, CVE-2026-20128 and CVE-2026-20122 with public proof-of-concept code (Cisco Talos, 2026-05-14). CISA added CVE-2026-20182 to its Known Exploited Vulnerabilities catalog on 2026-05-14 (CISA KEV catalog, 2026-05-14), under its existing Emergency Directive 26-03 for Cisco SD-WAN, first issued on 2026-02-25 (CISA ED 26-03).
Cisco lists no workaround; first fixed releases are 20.9.9.1, 20.12.5.4, 20.12.6.2 or 20.12.7.1, 20.15.4.4 or 20.15.5.2, 20.18.2.2 and 26.1.1.1, releases before 20.9 must migrate, and 20.10, 20.11, 20.13, 20.14 and 20.16 move to the next fixed train (Cisco PSIRT).
In May 2026, the Cisco Product Security Incident Response Team (PSIRT) became aware of limited exploitation of this vulnerability.
CISA's Emergency Directive 26-03 was first issued on 2026-02-25; on 2026-05-14 CISA added CVE-2026-20182 to its KEV catalog under it (CISA ED 26-03) (CISA KEV catalog, 2026-05-14). The SSH-key and NETCONF path Rapid7 demonstrated is on the SD-WAN Controller (Rapid7, 2026-05-14), and Cisco has since added the Validator to the affected products (Cisco PSIRT). Talos's ten other clusters exploit the separate SD-WAN Manager flaws CVE-2026-20133, CVE-2026-20128 and CVE-2026-20122, not this one (Cisco Talos, 2026-05-14). Cisco lists no workaround and no source proposes segmentation as one. Rapid7 demonstrates the bypass with its own Metasploit module, submitted to the public Metasploit Framework repository, so a working exploit is public (Rapid7, 2026-05-14).
CVE-2026-46300 ("Fragnesia") is a local privilege escalation flaw in the Linux kernel's XFRM ESP-in-TCP subsystem. Improper handling of shared page fragments during skb coalescing lets an unprivileged local user modify read-only file contents in the kernel page cache and gain root (Wiz Research, 2026-05-13). The public exploit uses user and network namespaces to obtain CAP_NET_ADMIN inside an isolated namespace, installs a crafted ESP security association over NETLINK_XFRM and overwrites the start of /usr/bin/su in the page cache only, so the on-disk binary never changes (Wiz Research, 2026-05-13). Dirty Frag discoverer Hyunwoo Kim says the flaw was accidentally activated by the patch for CVE-2026-43284 (Help Net Security, 2026-05-14). William Bowling of Zellic.io found it with the company's AI-agentic auditing tool, and a technical explainer and proof-of-concept exploit are public (Help Net Security, 2026-05-14).
The flaw is exploitable by unprivileged local attackers (Wiz Research, 2026-05-13), and CISA's catalog entry says it can be chained with the PHP2Shell flaws to implant PHP webshells and run commands as root (CISA KEV, 2026-10-28). Risk is highest on shared Linux hosts, container clusters where the page cache is shared across the host, CI runners and build farms, and cloud services running user code (Help Net Security, 2026-05-14). Ubuntu's default AppArmor restriction on unprivileged user namespaces is a partial mitigation that forces an extra bypass (Wiz Research, 2026-05-13).
Triage: unprivileged user namespaces are routine for rootless containers, sandboxed browsers and Flatpak, and IPsec is commonly used for VPNs and site-to-site tunnels (Red Hat RHSB-2026-003, 2026-07-03), so either alone is weak; the exploit's signature is one unprivileged process tree that sets up its own ESP security association inside a fresh network namespace and then runs su (Wiz Research, 2026-05-13).
Three vulnerabilities, collectively known as "Dirty Frag", have been identified in networking subsystems of the Linux kernel, two of which affect Red Hat products.
Red Hat says CVE-2026-43500 doesn't affect its products, while CVE-2026-43284 and CVE-2026-46300 do affect supported Red Hat Enterprise Linux kernels and anything built on them.
Researchers have already published working proof-of-concept exploits for Kubernetes, and Ubuntu spells out the risk of a container escape.
Red Hat rates CVE-2026-46300 CVSS 7.8 (Red Hat Product Security, 2026-05-13). Red Hat's own security bulletin groups it under the collective "Dirty Frag" name alongside CVE-2026-43284 and CVE-2026-43500, labelling it specifically "Fragnesia (skb coalescing via ESP-in-TCP)": "three vulnerabilities, collectively known as 'Dirty Frag', have been identified in networking subsystems of the Linux kernel, two of which affect Red Hat products" (Red Hat RHSB-2026-003, 2026-07-03). Applicability splits per CVE: "Red Hat says CVE-2026-43500 doesn't affect its products, while CVE-2026-43284 and CVE-2026-46300 do affect supported Red Hat Enterprise Linux kernels and anything built on them" (Aikido Security, 2026-09-04). Any RHEL, OpenShift or RHEL CoreOS estate must therefore confirm the CVE-2026-46300 fix specifically, not only the earlier two. For the Dirty Frag family as a whole (CVE-2026-43284, CVE-2026-43500 and CVE-2026-46300), Aikido writes: "Researchers have already published working proof-of-concept exploits for Kubernetes, and Ubuntu spells out the risk of a container escape"; it names no Kubernetes exploit for CVE-2026-46300 itself (Aikido Security, 2026-09-04). Because containers share the host kernel, a compromised workload that can reach the vulnerable code path (governed by seccomp, pod security and user-namespace settings) can escalate to root on the node (Aikido Security, 2026-09-04). As of 2026-09-05, no source stated confirmed in-the-wild exploitation of this specific CVE. Microsoft's previously reported limited in-the-wild activity ("privilege escalation involving 'su'") was itself stated as ambiguous between the "Dirty Frag" family and the separate, earlier "Copy Fail" vulnerability (CVE-2026-31431) (Microsoft Security Blog, 2026-05-08). That is not a distinction between the two Dirty Frag CVEs themselves, and not specific evidence of exploitation for CVE-2026-46300.
The interim mitigation is corrected. Until the kernel update is deployed, block and unload the esp4, esp6 and rxrpc modules where they are not needed (Wiz Research, 2026-05-13 · Help Net Security, 2026-05-14). Blocking esp4 and esp6 breaks IPsec, and Red Hat's IPsec-preserving alternative is disabling unprivileged user namespaces (Red Hat RHSB-2026-003, 2026-07-03). At disclosure AlmaLinux and CloudLinux had shipped patched kernels (Help Net Security, 2026-05-14). The entry previously named other kernel modules, a CAP_NET_ADMIN restriction, an affected version range and wider distribution patch coverage that its sources do not state, and it wrongly described Dirty Frag discoverer Hyunwoo Kim as a Wiz researcher.
Aikido's statement that working Kubernetes proof-of-concept exploits have been published concerns the Dirty Frag family as a whole, which it defines as CVE-2026-43284, CVE-2026-43500 and CVE-2026-46300, and its link for the container-escape risk is Ubuntu's page for CVE-2026-43284 (Aikido Security, 2026-09-04). No cited source describes a Kubernetes exploit for CVE-2026-46300 itself; its public exploit overwrites /usr/bin/su in the page cache (Help Net Security, 2026-05-14). On a host, that exploit shows as an unprivileged process creating user and network namespaces and installing an ESP security association over NETLINK_XFRM before su starts a root shell (Wiz Research, 2026-05-13).
CISA added CVE-2026-46300 to its Known Exploited Vulnerabilities catalog on 2026-10-28 and says in the entry that the flaw can be chained with CVE-2026-109618 and CVE-2026-109619 to implant PHP webshells and run commands as root, naming no actor or victim (CISA KEV, 2026-10-28). Dream Research describes a chain of this kind from artifacts its tooling recovered: a custom, non-public PHP2Shell build, tailored to the PAN-OS management web server, which runs PHP 8.0.30, gives code execution as the nobody user, and the Fragnesia exploit then overwrites /etc/pam.d/su in the page cache so that su root succeeds without a password; Dream says SELinux confines the web server's domain from modifying system files and that the whole chain runs in one HTTP request in under 15 minutes (Dream Research, 2026-10-29). From about 19,000 internet-facing PAN-OS firewalls that it found through a search engine, the attacker narrowed to 240 in the United States, Japan, Taiwan and South Korea, rooted 24 and planted a PHP webshell on each; on the second of two firewalls of a US government-affiliated organization (the first crashed) it set up an outbound SSH tunnel kept up by a systemd service and an rc.local entry, rooted the organization's firewall management server through the tunnel, decrypted the LDAP service-account password with root access and sprayed passwords against a domain that had no lockout threshold (Dream Research, 2026-10-29). Dream says the recovered artifacts confirm credential extraction and directory enumeration but could not tell whether the password spray succeeded, calls the campaign apparently Chinese and AI-assisted and does not attribute it to a named actor (Dream Research, 2026-10-27).
Dream Research updated its report on 2026-10-29 to say that the firewalls rooted with a custom PHP2Shell build chained to this kernel flaw run Palo Alto Networks PAN-OS, and counts 25 rooted PAN-OS firewall and management servers; Unit 42 says it independently verified the attacker's tooling and confirmed the reported results (Dream Research, 2026-10-29; Unit 42, 2026-10-29).
Dream Research's report, as updated on 2026-10-29, names the rooted firewalls as Palo Alto Networks PAN-OS and the exploited surface as the PAN-OS management web server, and says the recovered artifacts confirm credential extraction and directory enumeration in the attacked Active Directory, while whether the password spray succeeded is unknown (Dream Research, 2026-10-29). Unit 42 says the PHP flaws are exploited against PAN-OS management interfaces exposed to the internet and that customers who have applied the available PAN-OS fixes are not affected (Unit 42, 2026-10-29). CISA's catalog entry says the kernel flaw can be chained with the PHP flaws to implant PHP webshells and run commands as root; it does not itself say which flaw supplies the web-server code execution (CISA KEV, 2026-10-28).
CVE-2026-42945 (CVSS 4.0: 9.2, codename "NGINX Rift") is a heap buffer overflow present in src/http/ngx_http_script.c since NGINX 0.6.27 (2008) (depthfirst "NGINX Rift" technical writeup, 2026-05-13 · NCSC-CH Security Hub #12575, 2026-05-15). The root cause: when a rewrite directive combines an unnamed PCRE capture ($1, $2) with a replacement string containing ?, followed by another rewrite, if, or set directive in the same scope, ngx_http_script_start_args_code() sets a flag causing the write phase to URI-encode URI argument characters (expanding escapable characters such as + and & from one byte to three), while the length-calculation phase computed a shorter buffer without this escaping. The result is a deterministic out-of-bounds write into the adjacent heap allocation. F5 says the flaw may cause a heap buffer overflow in the NGINX worker process leading to a restart (ENISA EUVD, 2026-09-10). depthfirst's proof of concept achieves code execution with ASLR off, and depthfirst notes that forked workers share one heap layout, which could in theory be used to leak ASLR (depthfirst, 2026-05-13); F5's CVE description, updated on 2026-09-10, covers systems with ASLR disabled or where the attacker can bypass ASLR (ENISA EUVD, 2026-09-10). A working PoC is public at github.com/depthfirstdisclosures/nginx-rift. depthfirst's internal analysis system found the flaw on 2026-04-18; depthfirst reported it to NGINX on 2026-04-21, shared its RCE proof of concept on 2026-05-05, and F5 released the advisory on 2026-05-13 (depthfirst, 2026-05-13). NCSC-CH rated the exploitation status "UNKNOWN, Proof of Concept Available" on 2026-05-15 (NCSC-CH Security Hub #12575, 2026-05-15); VulnCheck's honeypot reports followed two days later (below). Temporary workaround: replace all unnamed captures with named captures (e.g., (?<user_id>...) → $user_id) in rewrite directives, which eliminates the vulnerable code path without requiring upgrade. Affected: NGINX Open Source 0.6.27 to 1.30.0 (nginx.org); NGINX Plus R32 to R36; Ingress Controller 3.5.0 to 3.7.2, 4.0.0 to 4.0.1 and 5.0.0 to 5.4.1; Gateway Fabric 1.3.0 to 1.6.2 and 2.0.0 to 2.5.1; Instance Manager 2.16.0 to 2.21.1; F5 WAF for NGINX 5.9.0 to 5.12.1; App Protect WAF 4.9.0 to 4.16.0 and 5.1.0 to 5.8.0; F5 DoS for NGINX 4.8.0; App Protect DoS 4.3.0 to 4.7.0 (Security Affairs, 2026-05). Fixed: NGINX Open Source 1.30.1 and 1.31.0 (nginx.org); NGINX Plus R32 P6 and R36 P4 (Security Affairs, 2026-05).
VulnCheck reported exploitation attempts against its honeypot networks; the nature of the activity and its goals are unknown (The Hacker News, 2026-05-17).
This vulnerability exists when the rewrite directive is followed by a rewrite, if, or set directive and an unnamed Perl-Compatible Regular Expression (PCRE) capture (for example, $1, $2) with a replacement string that includes a question mark (?).
The latest findings from VulnCheck show that threat actors have begun to weaponize the flaw, with exploitation attempts detected against its honeypot networks.
depthfirst reported the flaw to NGINX on 2026-04-21 and shared its code-execution proof of concept on 2026-05-05 (depthfirst, 2026-05-13); the earlier dates and the NDA had no source. VulnCheck reported exploitation attempts against its honeypots, with their purpose unknown (The Hacker News, 2026-05-17). The NGINX project rates the flaw medium (nginx.org), against the CVSS 4.0 score of 9.2 NCSC-CH gives (NCSC-CH Security Hub #12575, 2026-05-15).
F5's CVE description does not limit code execution to hosts with ASLR disabled. depthfirst's proof of concept runs with ASLR off, but depthfirst notes that forked NGINX workers share one heap layout, which could in theory be used to leak ASLR (depthfirst, 2026-05-13), and F5's CVE description, updated on 2026-09-10, covers systems with ASLR disabled or where the attacker can bypass ASLR (ENISA EUVD, 2026-09-10). An enabled ASLR is therefore no reason to defer the upgrade. Where the upgrade has to wait, depthfirst's workaround is to replace unnamed captures with named captures in every affected rewrite directive (depthfirst, NGINX Rift).
Composer validates configured GitHub tokens against an allowed character set in Composer\IO\BaseIO::loadConfiguration(); GitHub's new structured format for GitHub App installation tokens contains a hyphen, fails that check, and the resulting exception prints the full token to stderr, which CI captures in the job log, while GitHub Actions' secret masker may not redact it because the message is wrapped or framed (Packagist, 2026-05-13). Any workflow that configures a GitHub App installation token, including the auto-injected GITHUB_TOKEN, into Composer's auth and then runs any Composer command hits it, and actions such as shivammathur/setup-php register GITHUB_TOKEN automatically, so no unusual configuration is needed (Packagist, 2026-05-13). The advisory's header rates CVE-2026-45793 High 7.5 while its own text calls the severity medium, and it lists Composer from 2.3.0 before 2.9.8, from 2.0.0 before 2.2.28 and from 1.0 before 1.10.28 as affected, fixed in 2.9.8, 2.2.28 and 1.10.28 (Composer, 2026-05-13). CISA's Known Exploited Vulnerabilities catalogue did not list it as of 2026-10-21 (CISA KEV catalog, 2026-10-21).
The exposure window depends on the token: a hosted-runner GITHUB_TOKEN expires when the job ends or after six hours, and the error usually fails the job and expires it at once; on self-hosted runners it can stay valid for up to 24 hours after issuance; and tokens from actions/create-github-app-token or another GitHub App last up to an hour by default and can carry permissions broader than the workflow's own permissions: declaration (Packagist, 2026-05-13). GitHub rolled back the token change on 2026-05-13 at 14:30 UTC and planned to restart the rollout on 2026-05-18 after 14:00 UTC (Packagist, 2026-05-13).
The new format fails Composer’s validation, leading to an error message that exposes the full token contents to stderr.
Any workflow that configures a GitHub App installation token (including the workflow's auto-injected GITHUB_TOKEN) into Composer's auth and then runs any Composer command will hit it.
Usually the error leaking the token will also fail the job and immediately expire the token.
The leak was not tied to package-resolution errors: GitHub's new format for GitHub App installation tokens contains a hyphen, fails Composer's character validation when its configuration loads, and the exception prints the full token to stderr, so any workflow with such a token in Composer's auth hit it on the default path, including the auto-injected GITHUB_TOKEN (Packagist, 2026-05-13). Classic ghp_ personal access tokens are not affected by the validation bug itself, though the advisory notes the same leak applies to any future credential that fails validation (Composer, 2026-05-13). Packagist's relevant permission point is that App tokens can carry permissions broader than the workflow's own permissions: declaration, not that every GITHUB_TOKEN can write workflow files (Packagist, 2026-05-13).
The exposure was bounded: hosted-runner tokens usually expired with the failing job, self-hosted runner tokens stay valid for up to 24 hours and App tokens for up to an hour, and GitHub rolled the format change back on 2026-05-13 before planning a new rollout from 2026-05-18 (Packagist, 2026-05-13). Neither the Packagist post nor the advisory mentions Mini Shai-Hulud or exploitation; the advisory rates the flaw High 7.5, fixed in 2.9.8, 2.2.28 and 1.10.28 (Composer, 2026-05-13).
Sophos's State of Identity Security 2026, published on 2026-05-12, surveyed 5,000 IT and cybersecurity leaders across 17 countries and 14 industries: 71% reported at least one identity-related breach in the past 12 months, Switzerland reported the highest rate at 89%, followed by Mexico at 83%, and two-thirds of ransomware victims (67%) said their ransomware incident was directly tied to their most significant identity attack (Sophos, 2026-05-12). Weak management of non-human identities such as API keys and service accounts was a root cause in 41% of successful identity breaches (Sophos, 2026-05-12). Help Net Security reports that the energy, oil and gas, utilities and federal government sectors recorded the highest breach rates (Help Net Security, 2026-05-14).
Geographically, Switzerland reported the highest breach rate at 89%, followed by Mexico at 83%
Two-thirds of ransomware victims (67%) reported that their ransomware incident was directly tied to their most significant identity attack.
Sophos published the survey on 2026-05-12 (Sophos, 2026-05-12); 2026-05-14 is the date of Help Net Security's report (Help Net Security, 2026-05-14). Its ransomware finding is that two-thirds of ransomware victims (67%) said their ransomware incident was directly tied to their most significant identity attack (Sophos, 2026-05-12), and Sophos says attackers increasingly use stolen credentials as their primary mechanism for delivery (Sophos, 2026-05-12).
Patch CVE-2026-46300 on every Linux host that runs PHP or other internet-facing code, then on multi-tenant hosts, container nodes and CI runners: apply the distribution kernel update and reboot; until then, where IPsec is not in use, block and unload the esp4, esp6 and rxrpc modules (this breaks IPsec), and drop the page cache so a file already modified in memory reloads from disk.
On any PHP web host, appliance or management server that was reachable and unpatched for the PHP2Shell flaws, look for PHP files in web-reachable directories that nobody deployed, an outbound SSH session from the host that a systemd service or an rc.local entry keeps up, and password-spray bursts against Active Directory from the host's address, because Dream Research reports a chain of the two flaws that did exactly that.
2026-05-15-58b94fbd· Claude Sonnet 4.6 · window 36 h · 12 entries published
Items dropped (recency):
Kaspersky "State of Ransomware 2026" (Securelist, 2026-05-12): primary source is 72+ h before this run's start timestamp; outside the standard 36-hour window. The report qualifies for a dedicated treatment per PD-9 (annual/periodic threat report) but the Cisco SD-WAN/UAT-8616 chain is a higher-priority deep dive (criterion 1). Deferred to the next run for deep-dive treatment. out-of-window: primary source 2026-05-12, window_hours=36
Google "AI-developed zero-day for web admin tool" (BleepingComputer, 2026-05-11): primary source is outside the 36-hour and 72-hour windows. out-of-window: primary source 2026-05-11, window_hours=36
§ 4 Update not opened for CVE-2026-31431 "Copy Fail" KEV deadline: The CISA BOD 22-01 remediation deadline expiring today (2026-05-15) does not constitute material new development under PD-8, and is a US FCEB compliance date with no jurisdictional weight in Switzerland or the EU (PD-13). Original coverage: 2026-05-09. Operational signal is unchanged: kernel patches are available and should be applied.
Sub-agent telemetry: S1: returned (claude-sonnet-4-6, 447 s, 16 webfetch + 8 websearch + 4 bridge); S2: returned (claude-sonnet-4-6, 461 s, 9 webfetch + 18 websearch + 11 bridge); S3: returned (claude-sonnet-4-6, 881 s, 18 webfetch + 20 websearch + 2 bridge); S4: returned (claude-sonnet-4-6, telemetry not captured at context-summary time, content incorporated).
Fetch failures:
F5 K000161019 (primary vendor advisory for CVE-2026-42945): SPA gated behind myF5 customer portal login, my.f5.com returned only a loading error; advisory details sourced via researcher blog (depthfirst.com), NCSC-CH, and GitHub GHSA advisory. included with reduced confidence: F5 vendor advisory unreachable; researcher primary and NCSC-CH national-CERT advisory used.
CISA ED-26-03 directive body: Drupal page returned as raw HTML without extractable directive text body; directive details confirmed via Talos, Tenable, and SecurityWeek secondary sources. Source URL cited is the actual directive page (confirmed live).
inside-it-ch: Cloudflare Managed Challenge, no Swiss-specific items found via WebSearch fallback.
cert-eu: No new advisory above 2026-006 (2026-05-06) found in the 36-hour window.
Coverage gaps: advisories-ncsc-nl (no specific advisory ID obtained in this run); anssi-fr (no new CERTFR AVI for CVE-2026-20182 or CVE-2026-42945 identified; NCSC-CH and vendor primaries used); bsi-de (RSS fetched; WID-SEC-2026-1517 for Nextcloud used; no new SD-WAN-specific BSI advisory found).
Single-source items: CVE-2026-45793 (Composer token disclosure) is sourced solely from the Packagist vendor blog. No independent national-CERT advisory or security-researcher corroboration was identified in-run. The vendor blog is a credible primary source (Composer maintainers) and the disclosure URL is specific; however the claim has not been independently corroborated. Treat with commensurate confidence.
Verification status (cap-breach, 5 iterations, final verdict NEEDS_FIXES): Iter 1 (Opus): 3 truth, 1 editorial, Datadog item inverted, Fragnesia discoverer, UNC6780 alias, Nextcloud region. Iter 2 (Sonnet): 2 truth, 2 advisory, cluster attribution, §6 Nextcloud footer, Fragnesia co-discoverer, missed UAT-4356. Iter 3 (Opus): 3 truth, 1 editorial, Hyunwoo Kim regression, BitLocker unpatched count, "first time" claim, Composer framing. Iter 4 (Sonnet): 0 truth, 1 editorial, CVE table CVE-2026-45793 scope and patch column. Iter 5 (Opus, cap): 0 truth, 1 editorial, 5 broken internal anchor links in §6 (stale heading slug + -- vs - slugify pattern). All iter-5 findings remediated post-verification before publish. verification_residual_count: 1 (from final verifier).
S3 additional findings incorporated post-context-summary: CVE-2026-46300 "Fragnesia" (Wiz Research 2026-05-13, Help Net Security 2026-05-14, both live at fetch time); CVE-2026-45793 Composer token disclosure (Packagist blog 2026-05-13, live); Datadog Shai-Hulud framework open-source release (Datadog Security Labs 2026-05-13, live). All three URLs confirmed 200-OK in url-liveness.tsv.
Windows BitLocker YellowKey/GreenPlasma: No CVE has been assigned; sources are independent security researcher publications corroborated by NCSC-CH and multiple reputable news outlets. The BitLocker-bypass claim is corroborated by independent verification from Will Dormann and Kevin Beaumont (Mastodon/social media). GreenPlasma exploit chain reliability is lower (partial PoC, UAC prompt in default config), noted appropriately in § 1 item text.
CVE-2026-45691 Nextcloud: CVSS 5.9 (Moderate) does not clear a § 2 Trending Vulnerabilities gate. Included in § 1 due to high CH/EU public-sector relevance (Nextcloud is the dominant EU government self-hosted collaboration platform) and the identity/2FA bypass nature making it operationally relevant even without active exploitation.