View Ridge Security
Back to Cyber Hose

CISA adds exploited Zyxel GS1900 switch flaw CVE-2026-7273 to KEV

CISA added a Zyxel GS1900 switch vulnerability exploited by a suspected Chinese actor compromising nearly 1,000 devices worldwide.

๐Ÿšจ ACTIVE EXPLOITATION

  • CISA adds exploited Zyxel GS1900 switch flaw CVE-2026-7273 to KEV catalog
    CISA has added an actively exploited Zyxel GS1900 switch vulnerability (CVE-2026-7273) to its KEV catalog after a suspected Chinese actor compromised nearly 1,000 devices.
    • Applies to Zyxel GS1900 series Smart Managed Switches running firmware 2.90 versions and earlier, including GS1900-8, -8HP, -10HP, -16, -24, -24E, -24EP, -24HPv2, -48, and -48HPv2 models.
    • CVE-2026-7273 is a high-severity stack-based buffer overflow in a CGI program allowing unauthenticated LAN attackers to execute OS commands via crafted HTTP requests.
    • GreyNoise observed a suspected Chinese-speaking actor exploit the flaw and exfiltrate sensitive data from 996 switches across 48 countries as of September 17, 2026.
    • The campaign targeted over a dozen other vulnerabilities across a wide range of software and tech products.
    • Zyxel released patched firmware on June 16; CISA ordered FCEB agencies to patch by Thursday under BOD 26-04.
    Sources ๐Ÿ“Ž Coverage: bleepingcomputer.com ยท ๐Ÿ‘ via BleepingComputer

๐Ÿ’ฅ BREACHES & INCIDENTS

  • GHAPPIER supply chain attack poisons npm package and hits 65 GitHub repositories
    A supply chain attack tracked as GHAPPIER poisoned the npm package @dforge-core/dforge-mcp with a remote code loader and compromised 65 GitHub repositories.
    • Developers and organizations depending on the npm package @dforge-core/dforge-mcp were exposed; the attack reached 65 GitHub repositories, infecting 73 files across 22 accounts.
    • Malicious version 0.2.21 carried a hidden remote code loader; clean version 0.2.22 followed within about 36 minutes.
    • On September 9 an intruder used the package maintainer account for 105 minutes, inserting the loader and rewriting the release workflow for unattended publication.
    • The poisoned release carried valid npm provenance via GitHub Actions OIDC trusted publishing, with a Sigstore attestation recording the attacker's commit.
    • The loader, hidden in one line of a 99 KB file, launched a four-stage chain with a self-deleting final component; a linked PolinRider payload reads C2 data from Ethereum blockchain transactions and harvests cached Git credentials.
    Sources ๐Ÿ“„ Source: cloudsek.com ยท ๐Ÿ“Ž Coverage: cybersecuritynews.com ยท ๐Ÿ‘ via Cyber Security News, cryptika.com (discovered)

๐Ÿ”“ CVEs & KEV

  • CVE-2026-89422 โ€” Erlang OTP โ€” CVSS 9.3 โ€” TLS 1.3 client skips server authentication when ServerHello carries an unsoli...

  • CVE-2026-86296 โ€” D-Link DIR-822A โ€” CVSS 9.3 โ€” D-Link Router Hit by CVSS 10.0 Flaw Exploitable Remotely Without Authentication

  • CVE-2026-95511 โ€” Red Hat Enterprise Linux 10 โ€” CVSS 8.2 โ€” Cups: cups-filters: cups-filters: lpadmin can escalate to root via privileged...

  • CVE-2026-95508 โ€” Red Hat Enterprise Linux 10 โ€” CVSS 7.4 โ€” Libslirp: libslirp: heap buffer overflow in dhcpv6/tftp response builders on ...

  • CVE-2026-95503 โ€” Red Hat Build of Keycloak โ€” CVSS 6.8 โ€” Keycloak-services: keycloak-services: potential kdc spoofing bypass when kerb...

๐Ÿ•ต๏ธ RESEARCH & DEEP DIVES

  • SideCopy Expands India Targeting to Academia With ReverseRAT Spear-Phishing
    Pakistan-linked SideCopy is spear-phishing Indian academic institutions to deploy the ReverseRAT backdoor.

    • Targets: academic institutions in India, expanding beyond SideCopy's historical focus on Indian defense and government entities
    • Delivers a weaponized ZIP containing an LNK file disguised as "commskll.docx" with a spoofed PDF icon
    • LNK fetches an obfuscated HTA from "docsportal[.]in" and executes it via mshta.exe, reflectively loading a DLL dropper
    • Multi-stage in-memory XAML payload via .NET deserialization deploys ReverseRAT, which steals system metadata, screenshots, passwords, clipboard data and supports remote execution
    • C2 traffic encrypted with hard-coded key; data exfiltrated via port 5863 to "dns.educationportals[.]biz" (45.61.157[.]22)
    Sources ๐Ÿ“„ Source: trellix.com ยท ๐Ÿ“Ž Coverage: thehackernews.com ยท ๐Ÿ‘ via The Hacker News
  • SpyCloud: nearly 1 in 5 US water and wastewater organizations have active infostealer credential exposure
    A SpyCloud study found 1,787 of about 10,000 analyzed US water and wastewater organizations have credentials actively exposed by infostealers.

    • Applies to US water and wastewater utilities; SpyCloud analyzed 10,000 organizations from a database of 66,845 EPA-registered systems.
    • 1,787 organizations showed active infostealer exposure; 258 of them carried credentials to operational technology or remote-access systems.
    • One infected device at an unnamed smart meter provider held saved logins for roughly 167 US utility metering tenants, a cascading supply chain exposure.
    • Infostealer logs contain stolen session cookies, credentials, and autofill data that can bypass MFA via session hijacking; access brokers sell them to ransomware crews.
    • The study measured identity exposure, not OT devices or confirmed intrusion; SpyCloud began responsible disclosure with a CISA briefing.
    • Exposure concentrated in larger operators and the vendor supply chain, with small utilities largely underrepresented.
    Sources ๐Ÿ“Ž Coverage: cyberscoop.com ยท ๐Ÿ‘ via CyberScoop
  • Context Bombs Trick Autonomous Qwen AI Agents Into Stopping Cyberattacks
    Tracebit researchers used hidden 'context bomb' prompts in cloud decoys to make autonomous Qwen3.8-27B AI agents halt their simulated attacks.

    • Applies to autonomous AI hacking agents, tested on original Qwen3.8-27B and Blackfrost AI's 'abliterated' version in a controlled AWS environment.
    • Context bombs are defensive strings embedded in vulnerable resources like AWS Secrets Manager that agents encounter while scanning.
    • Earlier payloads triggering safety refusals failed; a new indirect prompt injection disguised as a valid operator command told the agent to end its assessment, and both Qwen versions complied.
    • Over 82 runs, the standard model gained admin privileges in 20.5% of 39 attempts versus 2.3% for the abliterated version, which was also slower (28.4โ€“29.9 min vs 13.5 min for first critical action).
    • Findings show abliteration (weight-level refusal removal) does not improve hacking effectiveness and highlight indirect prompt injection risk where hostile data is read as legitimate directives.
    Sources ๐Ÿ“„ Source: tracebit.com ยท ๐Ÿ“Ž Coverage: cybersecuritynews.com ยท ๐Ÿ‘ via Cyber Security News, cryptika.com (discovered)

๐Ÿ“‹ ADVISORIES

  • ๐Ÿ“„ Source for Suspected Chinese-speaking actor exploits wp2shell chain to steal 18,566 government records โ€” greynoise.io

Need help assessing your exposure?

Start with the free Posture Self-Check to see where you stand against the current threat landscape.

Free Posture Self-Check