← Carolina Clear Tech

Cyber Threat Brief

2026-10-24

Listen to this brief (14:55)

Download MP3
Show Notes

Show Notes - 2026-10-24

Stories Covered

CVEs Referenced

CVE-2024-57727, CVE-2025-64446, CVE-2026-0257, CVE-2026-109618, CVE-2026-109619, CVE-2026-110264, CVE-2026-110447, CVE-2026-112393, CVE-2026-112403, CVE-2026-112900, CVE-2026-114625, CVE-2026-115560, CVE-2026-35616, CVE-2026-41940, CVE-2026-97229

Indicators of Compromise

Domains: 254[.]43

IP Addresses: 5.2.3.18, 6.1.3.10, 6.4.2.12, 74.82.197.86, 38.54.50.151

Read the full brief

Get tomorrow's brief in your inbox

Protect Your Business

Need a security assessment? See our cybersecurity packages.

View Services

Today: Actively exploited Collabora CODE Server and Arista VeloCloud Orchestrator flaws have October 26 federal remediation deadlines; affected businesses should prioritize them now. Investigate LiteLLM 1.107.0 without launching Python or pip, and review ConnectWise-linked Entra application credentials for persistence. Patch PHP2Shell despite the distinction between widespread exploitation attempts and confirmed successful compromises.

Critical Alerts

Nextcloud CODE Server command injection under active exploitation (CVE-2026-110447)

Summary: CISA confirmed exploitation of Collabora Online Built-in CODE Server command injection, CVSS 9.2, critical. Within Nextcloud, the vulnerable proxy.php path allows remote code execution when the local CODE document server is not running; attackers have deployed web shells and searched for credentials, password-manager artifacts, and cryptocurrency wallets. Enrichment lists KEV remediation due October 26 and EPSS 1.2%, illustrating why observed exploitation outweighs a low predictive score.

Arista VCO authenticated SQL injection exploited (CVE-2026-114625)

Summary: Arista patched an actively exploited SQL injection in on-premises VeloCloud Orchestrator, CVSS 9.9, critical. An authenticated user, including one with Enterprise Read Only privileges, can acquire operator privileges and access internal functionality; hosted deployments are already patched. This is VCO's third exploited vulnerability since July, with KEV remediation due October 26 and EPSS 0.6%.

Ransomware & Extortion

INC and The Gentlemen affiliates use human-approved AI intrusion workflows

Summary: SOCRadar linked exposed operational records to affiliates of INC and The Gentlemen using an AI-executed, human-approved framework called Phantom. Records describe active exploitation of SimpleHelp CVE-2024-57727, FortiClient CVE-2026-35616, and cPanel/WHM CVE-2026-41940, followed by domain enumeration, credential theft, reverse-SOCKS exfiltration, and attempted security-tool impairment. Enrichment places all three in KEV, with EPSS scores of 96.8%, 94.0%, and 98.5%, respectively; other listed exploits were staged, developed, or referenced, not established compromises in this operation.

IOCs & Detection

Campaign-associated IP addresses

Summary: Arista identified two addresses associated with VCO exploitation, while SOCRadar identified an exposed server containing ransomware-affiliate operational records. These are campaign-specific leads, not proof of compromise from an isolated match; confidence below describes the strength of the source association.

Business & Infrastructure Threats

LiteLLM 1.107.0 compromise executes through Python startup hooks

Summary: Microsoft analyzed malicious LiteLLM 1.107.0 packages published October 21 through PyPI and official container images. Startup hooks could execute a credential collector without an application importing LiteLLM; observed Windows activity included activation while installing an unrelated dependency. Targets included developer, cloud, browser, SSH, and AI-service credentials, with additional payload retrieval supported on Unix systems.

Edge-device campaign harvests credentials and deploys remote-access tools

Summary: Group-IB and NyxLab attributed broad firewall and VPN targeting to k0t, with more than 70,000 exposed credentials associated with network devices and LDAP dumps. Observed methods included brute force and exploitation of FortiWeb CVE-2025-64446 and PAN-OS/Prisma Access CVE-2026-0257, followed by AD reconnaissance, remote-access persistence, and cloud exfiltration. Both vulnerabilities are in KEV, with EPSS 92.0% and 96.9%; researchers assessed possible ransomware connections but said the exact relationships remain unclear.

Aokigahara steals AI, developer, and cloud authentication material

Summary: Check Point observed Aokigahara collecting conventional browser, VPN, wallet, and password-manager data alongside AI application and development-tool files. Its targets include Claude, ChatGPT Desktop, Codex, Cursor, Windsurf, GitHub Copilot, Git credentials, SSH keys, and cloud configuration. Collection routines establish targeting, but not every targeted file contains a usable secret, and the researchers did not confirm monetization methods.

Windows / AD Security

Storm-0940 abuses ConnectWise application identities across Entra tenants

Summary: Microsoft attributed AI-assisted ConnectWise intrusions to China-linked Storm-0940, while Huntress observed access to hundreds of Microsoft 365 tenants, including more than 300 Entra tenants. Microsoft recorded over 151,000 failed logons and successful authentication through ConnectWise cloud-application identities, followed in some environments by Graph reconnaissance, application-credential changes, and privileged-role modifications. Added application credentials can preserve access after remediation of the originally affected application.

General Security News

Publisher support ended; Arc shutdown creates migration deadlines

Summary: Perpetual Microsoft Publisher support ended October 13, leaving installed copies without further security updates; Microsoft 365 removal updates start reaching Current Channel devices in mid-November. Separately, The Browser Company says Arc will become unusable on Windows and macOS December 22, with accounts deactivated and data deleted within 30 days. Both developments require application and data-retention planning.

Patch Priority

Vulnerability Disclosures

PHP2Shell patching remains urgent despite exploitation barriers (CVE-2026-109618, CVE-2026-109619, CVE-2026-115560)

Summary: PHP2Shell's two unauthenticated RCE vulnerabilities each carry CVSS v4.0 9.2, critical, with EPSS 11.6% and 6.0%. Researchers observed extensive probing and Mirai/Potassium installation attempts without confirmed successful wild compromises in these reports; laboratory results differ by deployment, so exploit difficulty is not a dependable defense. The separate CVE-2026-115560 requires an unusual runtime session-serializer change with upload progress enabled and is scheduled for a routine fix.

Ruby regex fixes address critical GitLab RCE (CVE-2026-97229, CVE-2026-110264)

Summary: Ruby fixed Onigmo memory-corruption bugs affecting applications that compile attacker-controlled regular expressions. GitLab's October CVEs both score 10.0, critical, and require no account; enrichment gives each EPSS 0.5%, while the article states neither was in KEV as of October 23. Developer-authored static patterns are not affected, and regex timeouts or pattern-length limits do not reliably prevent these bugs.

Pillow image-processing fixes require deployment-specific assessment (CVE-2026-112900, CVE-2026-112403, CVE-2026-112393)

Summary: Pillow 13.0.0 fixes EPS filename command execution and JPEG/TIFF memory-corruption vulnerabilities; GitHub CVE records rate these three 9.2, critical, although maintainer ratings differ. EPS exploitation requires attacker-controlled filenames and Ghostscript, while the JPEG-thumbnail and in-memory TIFF paths have different prerequisites and demonstrated impacts. Enrichment gives EPSS 0.4%, 0.3%, and 0.5%, respectively, and the article identifies no known attacks.

Trends & Context

AI-assisted campaigns still depend on familiar weaknesses: exposed management services, reused credentials, excessive application permissions, and persistent secrets. Today's response priority is to combine patching with identity investigation, since web shells, stolen tokens, and added OAuth credentials can survive remediation of the initial entry point.