←
SecureChap541 posts
SecureChap avatar

SecureChap

@SecureChap

Website security and pen testing!

◷ Joined March 2026▧ 8 Photos and videos
SecureChap avatar
SecureChap@SecureChap·6h

Seccomp on Telegram 13.2.2 (7303) only denies ptrace, bpf and mount while running in audit mode and letting every other syscall through. The first NebuSec bypass hits a QuickJS widget bug that lives outside the filter rules entirely, so no syscall check ever fires. The second chains a filtered widget bug to an unfiltered ioctl plus a kernel issue, landing straight into the calculator process without creating new programs. Widgets still share one address space. That single process is the only isolation left.

SecureChap avatar
SecureChap@SecureChap·10h

Keycloak drops to pre-auth RCE when a Base64 string carrying one extra '=' hits the AVX-512 VBMI decoder. The overflow writes 62-63 bytes past the output buffer in Java 17.0.6 and later, plus every 21/25/27 release. Java 8 and 11 stay clean except for the 8u511b33-perf pack. Default settings turn the bug into remote code execution on Keycloak, Jira, Confluence, Artifactory, Jenkins and PingFederate. Kafka, Solr, Elasticsearch, Gerrit and Spinnaker require prior auth. Oracle fixed it in 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. Corretto and Zulu carry the matching .1/.70 builds; Red Hat trees remain open. -XX:-UseBASE64Intrinsics

SecureChap avatar
SecureChap@SecureChap·22h

Unauth RCE is live against Oracle Java servers with zero creds required. Java 17 before 17.0.21.1 is affected, along with Java 21 before 21.0.13.1, Java 25 before 25.0.5.1 and Java 27 before 27.0.1.1. Java 8 and 11 stay untouched. Oracle shipped fixes 17 November 2026. Eclipse Temurin remains exposed on 17.0.21+9, 21.0.13 and 25.0.5+8. CVSS 4.0 score sits at 9.2. The flaw triggers through a remote code path in server deployments. Upgrade to the patched builds immediately.

SecureChap avatar
SecureChap@SecureChap·Nov 17

AVX-512 VBMI intrinsics inside java.util.Base64 can crash the JVM on decode for basic and URL modes. The path only activates on x86-64 CPUs that expose AVX-512 VBMI. Calls to Base64.getDecoder().decode() or getUrlDecoder().decode() reach the hot path once the JVM selects the optimized intrinsic. OpenJDK 17.0.6 and later ship the vulnerable code; Java 8 and 11 standard builds do not. Oracle 8u371-perf and later Enterprise Performance Packs added the same optimization. Fixes are in 8u513, 11.0.33.1, 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. Disable the affected decoder without a full upgrade: java -XX:-UseBASE64Intrinsics -XX:UseAVX=2 -jar yourapp.jar The same two flags also serve as a quick detection test on production hardware.

SecureChap avatar
SecureChap@SecureChap·Nov 17

GitDoor and GraphDoor abuse GitHub repos and SharePoint as covert C2 channels. **Initial access and loaders** MeshCentral Agent reached the hardcoded endpoint wss://27.102.66[.]162:443/agent.ashx after DLL sideloading. Legitimate executables (start.exe, pythonw.exe, zoomit.exe) loaded one of five malicious DLLs: jli.dll, SbieDll.dll, python39.dll, WINMM.dll or libcef.dll. Loader Type 1 ran RC4 followed by XOR; Loader Type 2 used XOR only. No specific CVEs or Windows build targets have surfaced. **GitDoor** The implant pulled tasks from two repositories under LatonyCharnleyaW. Havo was public, Trix private; both appeared at the end of 2024. Supported actions include keylogging to %APPDATA%\Microsoft\SystemCertificates\Cache\logs.dat, APC injection, cmd.exe execution, scheduled-task creation, registry enumeration, CookieKatz cookie theft, Mimikatz LSASS dumping and WhatsApp database key extraction. **GraphDoor** The implant created the Intel_Log directory under %APPDATA%, opened a named pipe, authenticated to login.microsoftonline.com:44… and staged data on SharePoint through Microsoft Graph API calls. Dev-platform C2 is convenient until defenders start watching the repos and tenants.

SecureChap avatar
SecureChap@SecureChap·Nov 16

Retadup's old USB vector is live again, dropping CoffeeLoader straight from the domains French authorities seized in 2019. An NSIS dropper writes a patched 7z.dll into %TEMP% and launches it via rundll32.exe. The Armoury packer (ArmouryAIOSDK.dll) now decrypts on the GPU by replacing OpenCL buffer calls with OpenGL texture uploads, then hands an intermediate shellcode downloader instead of a final DLL. The worm plants copies of 7z.dll inside 8-hex subfolders under %PROGRAMDATA% and %LOCALAPPDATA%, registers a Run key, and adds a second value that stops Task Manager from disabling autoruns. Shellcode is injected into dllhost.exe, svchost.exe or WerFault.exe before beaconing. Those same domains have served 32-bit Armoury-packed SmokeLoader since 2024 and switched to CoffeeLoader in early 2025; the loader accepts only three commands: process injection, save-and-run EXE, or save-and-run DLL via rundll32. Long-lived infrastructure lets operators swap payloads without rebuilding the worm.

SecureChap avatar
SecureChap@SecureChap·Nov 16

Anomalous nvidia-smi output on exposed Ray job agents flagged REF3364 operators deploying SANSAR in late October 2026. The Rust orchestrator stages under /opt/vadi_ and reaches its C2 via Arti over efxisnqldelyr5ra25oqfrsut54b34g26e7zrsiuupsoekjof2gho2id[.]onion. It first runs nvidia-smi to inventory GPUs, then drops kurt (XMRig 6.25.0) for CPU mining and tilki (SRBMiner-MULTI 3.6.9) for GPU. Persistence drops an SSH key into authorized_keys and an initramfs loader that installs acpi_thermal_mgr.ko. The rootkit builds on-target via DKMS. A modified NVML wrapper forces reported GPU utilization to 0 percent, temperature to 36-46°C, and power draw to 100-200W. The wrapper loads via LD_PRELOAD or by replacing libnvidia-ml.so in the library path ahead of any monitoring process. Hunt for unexpected LD_PRELOAD entries or mismatched NVML hashes on exposed Ray nodes.

SecureChap avatar
SecureChap@SecureChap·Nov 15

National IDs plus banking details give scammers everything needed for account takeovers and identity fraud in Taiwan. A sample of claimed TSMC employee data appeared on a hacker forum November 8. The poster alleges a larger set holding national IDs, addresses, contact info, and banking records. TSMC confirmed in an SEC filing that the data matches records stored by one of its cloud providers. The company said its own systems show no compromise and operations remain unaffected. The provider itself has not been named. Affected current and former employees are being notified directly. TSMC has brought in an outside incident-response firm. Staff should review bank and credit activity for anomalies, enable transaction alerts, and consider a credit freeze where available. One cloud partner just became the single point that exposed sensitive workforce data.

SecureChap avatar
SecureChap@SecureChap·Nov 15

21 N-central servers ended up running XMRig after a single payload drop from 157.20.241[.]16:8080/jup.sh. The entry point was CVE-2026-90436 in the bundled Apache Axis component. Unauthenticated attackers triggered unsafe JLine class instantiation during SOAP fault processing and got remote code execution as root. That foothold chained straight into an N-central privilege escalation bug. The combination yielded root on Linux and SYSTEM on Windows, with the miner configured to burn only half the cores. N-able shipped 2026.4 HF1 with tighter Agent and Probe paths, yet left the legacy API reachable in many deployments. Hotfix 2 did not close the Axis path when that API stayed enabled, and 19 of the 21 compromised boxes were already on HF2. Check legacy API status and confirm both components sit on the latest paired hotfix level.

SecureChap avatar
SecureChap@SecureChap·Nov 14

Cursor’s marketplace pushed malicious Claude Code extension versions 2.1.312 and 2.1.313 on 29 October after a stolen Fireworks AI key was used to pull traces from a read-only bucket. No CVE or GHSA has been assigned. The packages executed on install and wrote an attacker SSH key into ~/.ssh/authorized_keys, started sshd via launchctl on macOS or systemctl on Linux, then added a LaunchAgent or systemd unit for persistence. System enumeration ran through osascript and uname before phoning home. Fewer than 0.05 % of users had any coding traces exposed and fewer than 0.01 % had API keys taken. Cursor production, Privacy Mode data, the legitimate Anthropic extension, and Open VSX were untouched. Audit ~/.cursor/extensions for those two versions, remove the added SSH key, and rotate any credentials that may have been in the traces.

SecureChap avatar
SecureChap@SecureChap·Nov 14

CVE-2026-109618 and CVE-2026-109619 permitted unauthenticated RCE in PHP. PHP released patches on October 21. Exploitation began October 17. CISA added both to the KEV catalog on October 26. Attacks opened October 30 against Israeli organizations via the Check Point Security Gateway Mobile Access portal. A second wave hit multiple countries and sectors from November 2. Operators then used zero-day CVE-2026-123039 for root on the gateways, dropping PHP webshells, TSGA, and TSProxy. The gogo scanner located VMware ESXi hosts. The same infrastructure deployed an ESXi wiper. Internet-exposed gateways need explicit allow lists and monitoring to reduce supply-chain blast radius.

SecureChap avatar
SecureChap@SecureChap·Nov 14

A label tag carrying @label=;SQUIT target :injected crosses the server link untouched. The labeled-response module reads the incoming tag and writes it verbatim into the reply line sent over the server-to-server protocol. No sanitization occurs on the colon or newline delimiters, so the injected tokens are interpreted as a fresh command rather than a tag value. Versions 6.0.0 through 6.2.7 are affected. The maintainer shipped the correction in 6.2.7.1 on 7 November; active use of the flaw began the next day. CISA added the CVE to the KEV catalog on 14 November, setting a 16 November deadline for FCEB agencies under BOD 26-04. Operators unable to patch immediately can deploy the documented spamfilter workaround. Label tags must be escaped at the protocol boundary because the module treats their contents as opaque data while the receiving server treats the same bytes as executable commands.

SecureChap avatar
SecureChap@SecureChap·Nov 14

Gaia's /tmp/dynamic_socket is world-writable with zero authentication on the IPC channel. Any authenticated local user can connect and issue commands directly to dynamic_server, which blindly executes them as root. Active exploitation has run since late October, starting with PHP2Shell on the Mobile Access portal, then webshells, TSGA, TSProxy, and finally the ESXi wiper. Kill the process with `killall dynamic_server` until the hotfix is applied.

SecureChap avatar
SecureChap@SecureChap·Nov 13

XMRig miners landed on N-central Appliance 2026.4.1.26 HF2 after legacy API exposure. CVE-2026-90436 gave initial access through /dms/services/ServerUI and jline.internal.TerminalLineSettings once the legacy endpoint was enabled. VULN-2026-0009 then chained setLDFileOwner plus a symlink to turn a root-owned script writable by the service account, followed by serverControl invocation for root. Probes started 08:30 UTC; 157.20.241[.]16 delivered the 50 % CPU XMRig payload at 09:16 UTC, producing 21 active nc_ miners by 12:50 UTC, 19 on the November HF2 build. Disabling the legacy API (default off) breaks the chain at the first step. N-able advisory also pulled Intune integration; hosted instances stayed protected.

SecureChap avatar
SecureChap@SecureChap·Nov 12

Abdelhamid Naceri dropped SignDrop, abusing the CLFS .cnpf creation path when HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\CLFS\Authentication\Mode is set to 0 or 1. A low-privileged process triggers the logic and forces the CLFS engine to write the files under its own SYSTEM context. This sidesteps antimalware tamper protection and EDR self-defense without reusing any of the older memory corruption primitives. The same October 2025 mitigation that addressed prior CLFS authentication attacks is the exact setting being flipped. PoC code spawns a SYSTEM shell on Windows 11 26H2 and Windows Server 2025 Experimental builds. Toggle the mitigation back off and the older CLFS memory corruption paths reopen immediately.

SecureChap avatar
SecureChap@SecureChap·Nov 12

Concurrent HTTP/2 frames trigger a race in assign_headers() when mod_http2 processes headers from multiple streams at once. CVE-2026-119618 affects Apache HTTP Server 2.4.55–2.4.70 and mod_h2 2.0.8–2.0.42. No public exploits are known. Drop to HTTP/1.1 only while the module stays loaded: ``` <IfModule mod_http2.c> Protocols http/1.1 </IfModule> ``` The IfModule wrapper prevents the directive from breaking configs that do not load mod_http2 at all. Apply with: ``` apachectl configtest && systemctl reload httpd ``` Run `httpd -M | grep http2` and `ss -tlnp | grep :443` to confirm the module is either absent or restricted to HTTP/1.1 on every listening port.