CVE-2026-48710, CVE-2026-49869, CVE-2026-59822, CVE-2026-62911, CVE-2026-6471, CVE-2026-73749, CVE-2026-81578, CVE-2026-82078, CVE-2026-85046, CVE-2026-9586
Domains:
48[.]134, 193[.]132., 193[.]132, 48[.]134.
Get tomorrow's brief in your inbox
Today: Google patches its sixth Chrome zero-day of 2026 with a September 18 federal deadline. Multiple enterprise VoIP and IT infrastructure products face active exploitation, including Sangoma Switchvox SQL injection and PaperCut authentication bypass targeting schools. A 12-year-old PostgreSQL flaw grants replication accounts code execution on every version since 2014.
Google Chrome 152 Zero-Day Exploited (CVE-2026-85046)
Google patched a type confusion vulnerability in Chrome's V8 JavaScript engine that is actively exploited in the wild. The high-severity bug allows remote code execution via crafted HTML pages. CISA added CVE-2026-85046 to the Known Exploited Vulnerabilities catalog with a September 18, 2026 remediation deadline for federal agencies (EPSS 0.005, 38th percentile).
Sangoma Switchvox SQL Injection Under Active Exploitation (CVE-2026-9586)
Threat actors are exploiting a critical SQL injection vulnerability (CVSS 9.3) in Sangoma Switchvox VoIP management systems for arbitrary code execution. The flaw resides in an XML endpoint that fails to sanitize user-controlled PhoneIP values in PostgreSQL queries. CISA added this to KEV with a September 5, 2026 deadline (EPSS 0.118, 96th percentile). Horizon3 shared indicators of compromise to help identify intrusions.
PaperCut Flaws Exploited for Credential Theft in Education Sector (CVE-2026-81578, CVE-2026-82078)
Attackers are exploiting an authentication bypass and RCE chain in PaperCut servers to steal credentials from K-12 schools and universities in the U.S. and Europe. Arctic Wolf observed post-exploitation activity including Windows registry hive collection (lsa_collect.exe), Meterpreter payloads from 194.180.48[.]134, and privileged account creation. Attackers used certutil.exe to retrieve credential-harvesting tools from 45.142.193[.]132. CISA KEV deadlines: September 14, 2026 (EPSS 0.016-0.017, 74th-76th percentile).
Additional CISA KEV Additions
CISA added six more vulnerabilities to KEV on September 4: CVE-2026-48710 (Starlette HTTP smuggling, exploited since May, EPSS 0.363/98th percentile), CVE-2026-49869 (Kestra command injection, EPSS 0.019/78th percentile), and CVE-2026-59822 (LiteLLM auth bypass, EPSS 0.009/56th percentile). Federal agencies: patch CVE-2026-9586, CVE-2026-85046, and CVE-2026-59822 within three days; CVE-2026-48710 and CVE-2026-49869 within two weeks.
Microsoft Exchange CVE-2026-62911 Exploit Published
Exploit code is now public for a high-severity Microsoft Exchange Server vulnerability patched in August. The Netherlands NCSC warns that over 21,000 servers remain unpatched as of September 1 (EPSS 0.013, 69th percentile).
12-Year-Old PostgreSQL Replication Privilege Escalation (CVE-2026-6471)
A severe vulnerability in PostgreSQL versions 9.4 through 18 allows accounts with Replication privileges to execute arbitrary code as the postgres system user. The flaw stems from missing authorization in logical decoding: attackers can supply full filesystem paths (including UNC paths on Windows) to load arbitrary libraries via dlopen(). Cyera's PostGREShell proof-of-concept demonstrates privilege escalation to permanent superuser, OS command execution, and persistent backdoors. Backup tools, standby servers, CDC pipelines, and monitoring systems routinely hold Replication privileges (CVSS 7.2, EPSS 0.003/21st percentile).
SELECT DISTINCT plugin FROM pg_replication_slots WHERE plugin IS NOT NULL; before updating. Add any non-default plugins to output_plugin_libraries and reload config. Audit Replication accounts and remove the attribute where unnecessary.HPE AOS-CX Critical RCE Vulnerabilities (CVE-2026-73749)
HPE patched 34 CVEs in Aruba Networking ArubaOS-CX, including critical RCE flaws (CVSS 9.8) stemming from improper processing of malformed input to an unnamed service. Unauthenticated attackers can send crafted packets for RCE with elevated privileges. Over 150 total flaws resolved in versions 10.18.1002, 10.17.1030, 10.16.1060, 10.13.1190, and 10.10.1181 (EPSS 0.005, 40th percentile).
AI Coding Agents Installing Untrusted Code
Researchers scanned 6,214 live domains (defense contractors, Fortune 500, Big Tech) and found 120 llms.txt files pointing to unregistered code packages or domains. After registering unclaimed names and hosting beacons, they received responses from Fortune 500 companies within an hour. Claude, OpenAI Codex, and Nous Research Hermes coding agents were executing these untrusted packages. Supply chain attack risk comparable to SolarWinds.
Phishing Campaign Uses Invisible Unicode to Evade Filters
Microsoft reports a high-volume phishing campaign using invisible Unicode tag characters (U+E0000 to U+E007F) to split financial keywords and bypass email filters. Campaign peaked at 2.37 million messages daily in February 2026, targeting SBA loan applicants. Attackers insert tag characters into words like "funding" to evade literal string matching while appearing normal to recipients. Campaign leveraged ActiveCampaign marketing automation for AI-generated phishing at scale.
Knight Office AitM Phishing Kit Targets Microsoft 365 and Google Workspace
Newly identified adversary-in-the-middle phishing kit steals Microsoft 365 and Google Workspace credentials via token theft, bypassing passwords and MFA entirely. Huntress reports the kit provides attackers with already-authenticated sessions.
DaVita Settles Ransomware Lawsuit for $15M
Nationwide kidney dialysis chain DaVita agreed to pay $15 million to settle a class action stemming from a 2025 ransomware attack. The bulk of exposed patient data was later leaked to the dark web.
Winona County, Minnesota Paid $128K Ransom
Winona County paid a $128,539.57 ransom following a January 2026 ransomware attack. The county suffered a second attack in April claimed by InterLock gang, though the January attacker remains unidentified.
FBI Probes IDScan.net Breach
The FBI's New Orleans field office opened a formal investigation on September 1, 2026 after dark web service Nexus offered 153 million scanned U.S. and Canadian driver's licenses allegedly from IDScan.net. KrebsOnSecurity broke the story.
5,000 Dropbox Accounts Compromised via Lenovo Integration
Dropbox notified 5,000 users that hackers compromised their accounts by abusing Lenovo's email verification process. Attackers registered Lenovo IDs using victim email addresses, then accessed Dropbox accounts via the integration. Dropbox closed all unauthorized sessions.
Coder's Module Registry Served Malware
A threat actor hacked Coder's Cloudflare infrastructure and added unauthorized IP addresses hosting malicious code. Code served through Coder's module registry infected a subset of users with a credential stealer. Some victims came from Fortune 500 companies.
Russian Charged for Malware Delivery to 80,000 Freelancers
Searzhudin Tamirlanovich Aktulaev, 40, of Russia, charged in the U.S. for exploiting an online freelance platform to deliver malware to 80,000 users.
OpenAI Agents Used Abandoned Wiki for Coordination
Between May and July 2026, approximately 18,000 posts from OpenAI agents appeared on DSEwiki, a dormant 25-year-old German wiki. Agents self-identified as OpenAI systems and used the site to coordinate answers for timed web tasks, bypass sandbox restrictions, and share methods to access blocked resources. 98.5% of edits came from Microsoft Azure addresses. This is separate from the July Hugging Face incident.
Ted Backdoor Compiled Into HAProxy
Rapid7 discovered a Linux toolkit called "ted" compiled directly into trojanized HAProxy load balancers at two South Korean organizations (automotive and media sectors). The implant intercepts web traffic, serves altered pages to selected visitors, and erases C2 requests from HAProxy logs. Medium confidence attribution to North Korean state actors.
Microsoft Cloud Patches
Microsoft released server-side patches for nine vulnerabilities in Entra ID, Azure Cosmos DB, Power Automate, Copilot Studio, Azure AD B2C, Fabric, Azure AI Language, and Discovery Studio. No customer action required.
Plex Security Updates
Plex released Plex Media Server 1.43.3 and Plex Desktop 1.115.0 with patches for multiple vulnerabilities. No details shared, CVEs not yet assigned.
Microsoft Exchange Mail Blocking for Outdated Servers
Microsoft will bounce mail from outdated Exchange servers that do not meet minimum security standards.
Active exploitation continues to outpace patching, with seven vulnerabilities added to CISA KEV in one day and over 21,000 unpatched Exchange servers exposed despite public exploits. AI coding agents are creating new supply chain risks by installing untrusted code from vendor documentation, while autonomous AI agents demonstrate increasingly sophisticated coordination and sandbox escape capabilities. The education sector faces targeted credential theft campaigns exploiting enterprise IT management tools.