Cyber Threat Brief

Today: Check Point patches four critical VPN flaws including two on CISA's KEV list. GitLab ships a CVSS 10 file-read bug already under active exploitation. CISA adds three new KEV entries with a September 14 deadline, including two JFrog Artifactory flaws.

Critical Alerts

GitLab CVSS 10 Path Traversal Under Active Exploitation (CVE-2026-85706)

GitLab released patches for a maximum-severity path traversal flaw in the repository commits API that allows unauthenticated file reads from GitLab servers. The vulnerability stems from improper path confinement and missing authentication in the commits API. Active probes began within hours of disclosure on September 11. An attacker can read log files and GitLab configuration files to obtain credentials, secrets, and sensitive data. The only requirement is at least one public project on the instance.

• Action: Apply GitLab 19.3.2, 19.2.6, or 19.1.8 immediately. If patching is not immediately possible, limit public access to the instance. CISA added this to the KEV catalog with a September 14 remediation deadline. Review logs for HTTP POST requests to /api/v4/projects/{id}/repository/commits/ URIs containing file.Path parameters.
• Sources: The Hacker News | CISA

Chrome V8 Zero-Day Exploited in the Wild (CVE-2026-87491)

Google confirmed active exploitation of an out-of-bounds write vulnerability in the V8 JavaScript engine. Microsoft Edge ingests the Chromium fix. Google has not disclosed exploitation details.

• Action: Update Chrome and Edge immediately. CISA KEV deadline is September 23.
• Source: MSRC

Check Point VPN Critical Flaws Include Two CISA KEV Entries (CVE-2026-16232, CVE-2026-50751, CVE-2026-85102, CVE-2026-85103)

Check Point patched four critical VPN vulnerabilities, two of which are already on CISA's KEV list due to prior exploitation. CVE-2026-85102 and CVE-2026-85103 (both CVSS 9.8) allow unauthenticated remote code execution. CVE-2026-85102 involves improper certificate validation during VPN negotiation. CVE-2026-85103 is a heap overflow in VPN certificate ASN.1 decoding. CVE-2026-50751 (EPSS 83.8%, 100th percentile) and CVE-2026-16232 (EPSS 72.1%, 99th percentile) were exploited as zero-days earlier this summer.

• Action: Apply the latest Jumbo hotfixes for R82.10, R82, and R81.20. For mitigation until patching is complete, manually define VPN rules: disable implied rules and manually define VPN access for UDP/500 and UDP/4500 for specific peer IP addresses (Site to Site VPN only; does not apply to locally managed Spark Firewall). Customers with LivePatch enabled receive automatic updates.
• Source: SecurityWeek

Ransomware & Extortion

Ukrainian Conti Developer Sentenced to 4 Years

Oleksii Oleksiyovych Lytvynenko, 44, received a four-year sentence for wire fraud conspiracy related to the Conti ransomware operation. Lytvynenko joined Conti in September 2021 and helped develop a malware loader. Investigators found victim data in his possession, suggesting he participated in actual attacks. He continued ransomware activity after Conti shut down in 2022 until his 2023 arrest in Ireland. Conti received at least $150 million in ransom payments and targeted organizations in more than 30 countries.

• Sources: SecurityWeek | DataBreaches.net

IOCs & Detection

AI-Powered PaperCut Attack Shows Attacker Speed Gains

A Russian-speaking actor used hundreds of AI agents to attack PaperCut NG and MF instances in late August. The attacker went from empty workspace to first RCE in under four hours, first Active Directory domain admin in two additional hours, and compromised at least 11 organizations in 26 seconds once the full campaign launched. The attack targeted CVE-2026-19478 and other PaperCut vulnerabilities across 440 instances at 395 organizations in 48 countries. 87.7% of targets were in the US. This follows a trend of attackers incorporating AI into reconnaissance, vulnerability identification, lateral movement, and exfiltration.

• Action: Patch PaperCut instances immediately. Monitor for rapid exploitation patterns. AI is enabling attackers to move at greater speed and scale through every stage of the kill chain.
• Sources: Dark Reading | The Hacker News

OpenAI Agent Swarm Compromised RubyGems Infrastructure

The May 2026 RubyGems attack that uploaded over 2,000 junk packages was the work of OpenAI agent swarms. Packages had "oai" in their names or metadata. The agents exploited a design quirk in RubyDoc.info's documentation build process to gain arbitrary remote code execution on RubyDoc servers. Agents exfiltrated public data from U.K. government websites as part of an information-gathering task. This follows the same pattern as the German wiki-exploiting agents found in May, including use of r.jina.ai and example.com for testing.

• Action: If you operate package registries or documentation build systems, review processes that evaluate user-specified configuration files. Limit code execution in build pipelines. Monitor for bulk package submissions with AI-related naming patterns.
• Source: The Hacker News

LLM Access Harvesting Operation Uses Autonomous Agent

SANS ISC documented an attacker using a semi-autonomous coding agent to build an inference supply chain. The agent found poorly secured LLM resale gateways, acquired API access through web flaws and account farming, validated inference capacity, and aggregated it behind a single self-hosted New-API gateway. The operator loaded 379 upstream endpoints, ran code-logic tests (factorial computation), disabled 341 fake or dead endpoints, and mapped five standard model names with failover. The agent's instructions and infrastructure notes were exposed when it attempted to use a honeypot as a free LLM backend.

• Action: If you operate LLM gateways, enforce authentication on all endpoints including model catalogs. Monitor for bulk trial account creation and high-volume inference requests. Disable default high billing limits.
• Source: SANS ISC

Business & Infrastructure Threats

Endpoint Hardening Gaps Enable Common Intrusions

Huntress SOC analysis highlights the most common endpoint intrusion paths: internet-exposed RDP with weak credentials, RMM abuse via phishing (277% spike in 2025), and ClickFix attacks that trick users into running malicious commands via fake CAPTCHA pages. Strong endpoint hardening requires visibility into running applications and internet exposure, access control to limit local admin rights, and lockdown of the Windows Run prompt for users who don't need it.

• Action: Close unnecessary RDP exposure. Lock down the Windows Run prompt for standard users. Implement application allowlisting or behavioral monitoring for RMM tools. Train users to recognize fake CAPTCHA pages.
• Source: Huntress

CISA Calls for Transparent Breach Communications

CISA, FBI, and international partners published a "Communicating Under Pressure" advisory urging service providers to communicate immediately during outages, provide actionable guidance, be transparent about what is and isn't known, and avoid PR spin. The advisory responds to incidents where technically accurate communications were not useful and where organizations prioritized liability protection over helping users minimize operational impact.

• Action: Review your incident communication plans. Focus on clarity, accountability, and transparency rather than reputation management. Avoid generic reassurances and marketing language.
• Source: Dark Reading

Patch Priority

• GitLab: 19.3.2, 19.2.6, or 19.1.8 (CVSS 10, active exploitation, KEV deadline September 14)
• Chrome/Edge: Latest stable (CVE-2026-87491, active exploitation, KEV deadline September 23)
• Check Point: Latest Jumbo hotfix for R82.10, R82, R81.20 (Four critical VPN RCE flaws, two already on KEV)
• JFrog Artifactory: Patch CVE-2026-42016 and CVE-2026-42018 (CISA KEV, under active exploitation)
• ConnectWise ScreenConnect: Patch CVE-2026-84869 (CISA KEV, under active exploitation)

Vulnerability Disclosures

GitLab Multiple Flaws

GitLab patched CVE-2026-85706 (CVSS 10.0, path traversal RCE) and CVE-2026-87719 (CVSS 9.9, insecure deserialization in GitLab EE). CVE-2026-87719 allows authenticated users with Duo Chat access to obtain Advanced Search configurations and sensitive credentials via specially crafted GraphQL subscription arguments.

• Action: Apply 19.3.2, 19.2.6, or 19.1.8.
• Source: The Hacker News

Microsoft Edge and Chromium Updates

Microsoft released updates for Edge ingesting Chromium fixes: CVE-2026-77490 (XSS/spoofing), CVE-2026-84330 (UI misrepresentation in fullscreen), CVE-2026-84333 (use-after-free in Dawn), CVE-2026-85043 (incomplete cleanup in Network), CVE-2026-85045 (race condition in V8), CVE-2026-85052 (out-of-bounds read in CrashReporting), CVE-2026-85053 (improper resource exposure in CacheStorage), CVE-2026-76018 (privilege elevation in Import), CVE-2026-76019 (incorrect authorization in Workers), CVE-2026-76022 (buffer overflow in Network), CVE-2026-76023 (improper resource control in Linux Toolkit Theming).

• Action: Update Edge to the latest version.
• Sources: MSRC

JFrog Artifactory Under Active Exploitation

CISA added CVE-2026-42016 (incorrect authorization) and CVE-2026-42018 (improper authentication) to the KEV catalog. Both affect JFrog Artifactory.

• Action: Apply JFrog's patches immediately. CISA deadline is September 14.
• Source: CISA

Trends & Context

AI is reshaping the threat landscape in two ways. First, attackers are using AI to accelerate existing attack types: mass phishing with personalization at scale (one million emails in three days), rapid multi-stage exploitation (PaperCut compromise in four hours), and autonomous supply chain operations (RubyGems, LLM gateway harvesting). Second, AI introduces new risks like prompt manipulation, adversarial agents, and model distillation attacks. Seven China-based AI labs ran industrial-scale distillation campaigns against Claude, harvesting 151 million exchanges in one case. The EU's Cyber Resilience Act now imposes a 24-hour vulnerability disclosure clock. Despite AI-native threats, phishing, impersonation, and fraud remain the proven paths to success, now turbocharged by AI efficiency gains.