CVE-2022-21894, CVE-2026-109618, CVE-2026-109619, CVE-2026-110264, CVE-2026-111497, CVE-2026-111943
Domains:
litellm[.]dev, 8[.]57, 201[.]8, 220[.]92, 184[.]166, 200[.]5, 159[.]64, 163[.]188, 98[.]37
Get tomorrow's brief in your inbox
Today: Patch PHP2Shell immediately. Treat affected LiteLLM installations as a credential-exposure incident and avoid starting Python during initial containment. ConnectWise customers should keep affected connections disabled until formal recovery instructions apply to their service.
PHP2Shell exposes web-facing PHP to unauthenticated code execution (CVE-2026-109618, CVE-2026-109619)
Summary: PHP's two vulnerabilities are rated CVSS v4.0 9.2, Critical. The PHP project identifies all installations serving web requests as at risk, including PHP-FPM and Apache mod_php; mod_php has a demonstrated attack before application code runs. Amazon observed an October 17 exploitation attempt against its sensors, before public disclosure, while cautioning that success against real systems was unclear.
Malicious LiteLLM 1.107.0 requires incident response
Summary: LiteLLM says its October 21 release compromise distributed credential-stealing and remote-code-loading malware in 1.107.0 wheels and container images, including latest, main-latest, and main-stable during the incident window. The official GitHub source tag is clean; artifact provenance matters.
PingCAP confirms copied business data; customer-specific exposure remains under investigation
Summary: PingCAP's October 19 notification says stolen service-account credentials enabled access between September 9 and October 7, 2026, with data and source code copied. Exported records included business and billing information, SQL editor text and errors, and encrypted API-key values; not every customer necessarily had every category exposed. One investigator-owned encrypted key value could not be used directly. Whether other usable credentials were exposed remains unresolved. Investigators found no evidence of command execution or persistence in reviewed production systems, or impact to data stored in TiDB; these findings do not negate the confirmed exposure.
Confidence here describes the source's association with this incident. A matching address or filename alone does not establish compromise of a particular system.
LiteLLM credential theft and subsequent cloud activity
Summary: Wiz identified an attacker-controlled exfiltration destination and observed stolen AWS credentials being used from two source addresses.
litellm[.]dev, including www.litellm[.]dev/healthcheck. Recommended: block the malicious lookalike destination and search historical DNS/proxy records.cook.pth, cook.py, sitecustomize.py, and /tmp/.pc_once_72129118c59f5636c0bb. Wiz notes the marker can fail on Windows. Recommended: preserve matching files for analysis; absence of a marker does not clear an affected installation.168.100.8[.]57 and 165.227.201[.]8. Wiz observed AWS credential validation and subsequent discovery from these addresses. Recommended: correlate with sts:GetCallerIdentity, bedrock:ListFoundationModels, and unusual IAM discovery rather than treating an isolated IP match as proof.PHP2Shell historical hunting leads
Summary: Amazon lists 103.216.220[.]92, 31.223.184[.]166, 37.19.200[.]5, 134.209.159[.]64, 103.114.163[.]188, and 95.217.98[.]37 in its exploitation telemetry. These include shared anonymization infrastructure. Amazon recommends investigation rather than blocking these addresses solely on this association.
Storm-0940 abuses ConnectWise application trust in Microsoft 365
Summary: Microsoft attributes the ConnectWise multi-tenant application compromise to China-state-aligned Storm-0940 and describes its use of ARTEX for automated attacks. Compromised application credentials enabled Microsoft 365 authentication without interactive user sign-in. Application presence or successful authentication alone does not establish data access or exfiltration. ConnectWise's October 22, 13:00 UTC update says phased restoration is progressing; existing backup archives remain encrypted, isolated and preserved, with no indication of compromise. Individual tenants still require review.
BlackX bootkit targets systems where attackers already hold administrator access (CVE-2022-21894)
Summary: Kaspersky describes BlackX in attacks against diplomatic agencies in Central Asia. Attackers installed it after compromising networks and obtaining administrator privileges. Its BlackLotus-like exploitation of CVE-2022-21894 manipulates Secure Boot through custom Machine Owner Keys; a subsequent driver interferes with security callbacks and ETW providers.
Bronze Sandcastle uses legitimate executables to load malicious DLLs
Summary: Sophos observed legitimate Sandboxie executables loading malicious sbiedll.dll files in multiple victim environments. Sideloaded DLLs read encrypted config.ini payloads and execute code in memory; the decrypted payload is a backdoor.
SandboxieWUAU.exe loading unexpected sbiedll.dll files, especially beside encrypted configuration files in unusual directories. Correlate module loads, file reads and execution paths. A legitimate filename alone is insufficient grounds for a compromise verdict.Chrome adds ClickFix command-copy warnings for Enhanced Protection users
Summary: Google says Chrome detects suspicious commands copied by Enhanced Protection users and sends the payload, after removing personal information, to Safe Browsing for analysis. A dangerous verdict triggers a warning. This targets social engineering that persuades users to execute commands outside the browser.
GitLab fixes unauthenticated regex-engine RCE (CVE-2026-110264)
Summary: GitLab rates this issue CVSS 10.0 and says it can permit unauthenticated remote code execution under certain conditions. The October 21 release fixes affected CE/EE branches with 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10, and 18.11.13. GitLab.com is patched; GitLab Dedicated builds 19.3.5-internal0 and 19.4.2-internal0 or later already include the fixes.
Go rejects compromised eMudhra intermediates (CVE-2026-111943)
Summary: Talos describes eight weak intermediate certificates: six for TLS and two for code signing. It reports that the relevant browser/platform vendors blocked them and eMudhra revoked them and stopped issuing under its G3 root. Go's own advisory identifies affected crypto/x509 versions before 1.26.10, and the 1.27 release line before 1.27.3. The fix rejects the known intermediates during certificate verification.
Node.js fixes Brotli decoder RCE (CVE-2026-111497)
Summary: Node.js's October 21 security release addresses ten issues. It rates the bundled Brotli decoder out-of-bounds write High, identifies possible remote code execution, and lists 22.x, 24.x and 26.x as affected release lines.
Recommended priorities: inventory shared runtimes and release artifacts, confirm which application paths are exposed, and investigate unauthorized changes to privileged application identities. For a malicious package, follow the vendor's compromise assumptions rather than waiting for a visible payload or confirmed theft. Recovery should address exposed credentials and persistence as well as replacing affected software.