Daily ยท 2026-09-11
๐ฏ Check Point VPN 9.8 RCE, ๐งต JFrog Artifactory admin-token chain, ๐ค Claude agent campaigns, ๐ก๏ธ Defender ShieldCrash fixed, ๐ฐ๏ธ UNC3569 Sogou backdoor, ๐๏ธ IDScan 153M licenses
Check Point patched two CVSS 9.8 unauthenticated RCEs in its VPN gateways and Wiz detailed the full three-CVE Artifactory chain, while Anthropic and Google documented AI agent frameworks running intrusions at machine speed. A signed Sogou update channel dropping GRAYRABBIT, a 153-million-record IDScan breach, new KEV additions for MikroTik and N-central, and fresh IoT and RAT malware round out the day.
Check Point has patched two CVSS 9.8 flaws in its VPN stack, and the pair is the day's most consequential new perimeter exposure. Check Point VPN RCE: CVE-2026-85102 combines an authentication bypass with remote code execution while CVE-2026-85103 is a separate RCE in certificate handling, and both affect Remote Access VPN and Site-to-Site VPN on Check Point security gateways. The company's internal research team found and fixed the bugs and reports no exploitation yet, but an unauthenticated RCE on an internet-facing gateway is a direct route to lateral movement, credential theft and ransomware staging, so patching should be treated as urgent.
Wiz has reconstructed the full chain behind the Artifactory intrusions that were first reported as a single flaw. JFrog Artifactory exploitation chained CVE-2026-42018, which leaks an anonymous-user JWT even when anonymous access is disabled, into CVE-2026-42016, a token scope-validation flaw, to mint administrator credentials, with CVE-2026-82329 as the underlying authentication bypass. Wiz observed the activity against self-hosted deployments between August 15 and September 8, using a POST to /access/api/v1/aws/token/ and then /access/api/v1/tokens, and admin access lets attackers replace build artifacts, steal repository credentials and deploy server-side plugins. That turns the single-CVE exploitation reported earlier this month into a repeatable three-step chain, so self-hosted operators should rotate tokens and audit artifacts and plugins they did not publish.
Two disclosures make the same point about agentic offense. Anthropic's September misuse report covers December 2025 to August 2026 and describes a Russian-aligned campaign against more than 20 government and defense organizations in Ukraine and Europe, two Chinese undergraduates who produced more than a dozen candidate zero-days in a single month, and ShinyHunters affiliates dumping 2,100 cloud access tokens across 40 tenants in 34 hours. Google and Mandiant's multi-agent campaign describes a financially motivated actor using an AI coding chatbot with Markdown playbooks to scan, troubleshoot and rotate infrastructure while running a credential-harvesting operation end to end in under six hours. The common thread is that AI now supplies the orchestration layer across reconnaissance, exploitation and exfiltration, lowering the skill floor for campaigns that used to require a team.
Microsoft's September cumulative updates for Windows 10 and Windows 11, KB5122878, KB5124008 and KB5122880, now carry the fix for the ShieldCrash Defender zero-day, which follow-on reporting describes as an exploited flaw in the Malware Protection Engine that yields SYSTEM on the host. It is the third consecutive bypass of the same component this year, after RoguePlanet and ShieldBreak, and Defender engine deployment should be confirmed at the fixed build rather than assumed.
KATARU is a new IoT malware family that brute-forces Telnet, drops an ARM payload called vlxx.arm through BusyBox, then tries three public Linux privilege-escalation exploits, CVE-2026-46300 (Fragnesia), CVE-2026-43284 (DirtyFrag) and CVE-2026-31431 (Copy Fail), plus a cgroup v1 release_agent escape and a direct /etc/passwd root-password wipe. Researchers caught it after a honeypot was brute-forced from 160.191.242.92, and unlike typical Mirai variants it adds encrypted C2, anti-analysis checks and decoy traffic. Compromised devices become DDoS infrastructure, so patch the three kernel bugs and close Telnet on embedded systems.
IBM disclosed two CVSS 9.6 flaws in DataStage on Cloud Pak for Data version 5.4.0.0. CVE-2026-82107 is an improper-authentication bug that lets a remote authenticated attacker pull sensitive information, and CVE-2026-82100 is a path-traversal flaw that can force a denial of service. Both are reachable over the network with low privileges and no user interaction, and neither has a public proof of concept or reported exploitation yet, so patch through IBM's advisory before one appears.
China-linked UNC3569 used the update mechanism of Sogou Input Method to deliver the GRAYRABBIT backdoor, a signed supply-chain path that bypasses network defenses because the install runs with elevated privileges and the payload carries vendor trust. Sogou Input Method has hundreds of millions of users in China and UNC3569 has historically targeted government and defense organizations, so the delivery vector matters more than the malware itself. Audit Sogou installs, verify update integrity and treat application whitelisting as the practical control.
IDScan.net has confirmed that its cloud environment was breached and a database of 153 million driver's license scans was exposed, including names, addresses, dates of birth, license numbers and in many states facial photographs. The scale covers a large share of the US driving-age population and the data is well suited to identity theft and synthetic identity fraud. Organizations that use IDScan for verification should assume their customers are in the exposed set and plan notifications, and the incident is another argument against retaining identity documents at scale.
N-able shipped a fourth emergency hotfix in five weeks for CVE-2026-86218, a CVSS 10.0 unauthenticated RCE in N-central that is now on CISA's KEV list, after Huntress found an already-patched customer had been compromised. N-central is the management plane MSPs use to administer client estates, so a pre-auth RCE there can become a supply-chain event across thousands of downstream networks. Patch the current hotfix, then hunt for rogue accounts, tunnels and script pushes.
CISA added MikroTik RouterOS CVE-2026-67277 and CVE-2026-86060 to the KEV catalog on September 10, confirming the RouterOS vulnerabilities as exploited, with CVE-2026-67276 a signature-verification bypass, CVE-2026-67277 a missing-authentication flaw and CVE-2026-86060 an argument injection for privilege escalation. MikroTik patched them on September 3 across RouterOS 6.x before 6.49.21, 7.x Long-Term before 7.23.4, 7.x Stable before 7.24.2 and the development branch before 7.25 beta 3. CERT Polska confirmed in-the-wild use, so patch any RouterOS device with SSH exposed and check logs for the Flagged marker.
Hunt.io has tied a July breach at King's Lynn and West Norfolk Borough Council to what it calls a mass campaign against SonicWall SMA1000 appliances exploiting CVE-2026-15409, which was disclosed only two days before exploitation began. The operator left a toolkit in an open directory showing roughly 250 targets and credentials stolen from at least nine Active Directory domains, with confirmed victims in France, India, Italy and the United States. The credential theft ran from the edge appliances themselves, so audit SMA1000 patch status and authentication logs, and expect the stolen credentials to resurface.
GeoVision's firmware 1.14, released September 10, fixes 23 vulnerabilities across the GV-LPC2011 and GV-LPC2211 license plate recognition cameras, led by CVE-2026-88285, a CVSS 9.4 authentication bypass, plus CVE-2026-88271 privilege escalation to administrator, CVE-2026-88277 OS command injection and CVE-2026-88289 buffer overflow. The documented chain lets a low-privilege guest pull plaintext admin credentials from the SSVR video interface, overwrite the admin password and reach root through unsanitized shell parameters. No exploitation is confirmed yet, so update from 1.13 or earlier, isolate cameras on a surveillance VLAN and stop exposing their management interfaces.
Zscaler ThreatLabz has detailed SloppyRAT, a Windows remote access trojan built to enable ransomware intrusions. It arrives through a ClickFix chain that abuses the legitimate finger.exe utility on TCP port 79 to fetch a batch script from finger.linked4x[.]com, copies and renames curl.exe, then stages the RAT, which combines encrypted code, indirect system calls, certificate pinning and a blockchain fallback for locating its command-and-control. Capabilities include reconnaissance, remote command execution, reverse proxying and lateral movement, so block outbound TCP 79 and restrict finger.exe execution.
Researchers have described PuzzleMask, a prompt-obfuscation technique that hides policy-violating instructions inside ordinary English prose so a fast screening model passes them while a stronger downstream model can still recover and act on them. It is not a jailbreak by itself, but combined with one it defeats the two-stage gatekeeper design many AI applications rely on, because the gap in reasoning depth between screener and target is the actual weakness. Deployments that use a cheaper model as a safety pre-filter should assume it can be talked past.
The Canadian Centre for Cyber Security's WebPros advisory AV26-908 covers a cPanel EmailTrack SQL injection, CVE-2026-67401, and two critical remote-code-execution flaws in the ConfigServer Firewall plugin, CVE-2026-65638 and CVE-2026-65639, affecting CSF versions 14.00 through 16.29. Because CSF sits alongside cPanel on a large share of self-managed hosting servers, the combined attack surface is significant. Apply WebPros' security releases and update cPanel builds below 11.110.0.143, 11.134.0.55, 11.136.0.39, 11.138.0.4 and WP2 11.138.1.9.
PaperCut has replaced its emergency patches with full fixes for the two actively exploited print-management flaws, CVE-2026-81578 and CVE-2026-82078, according to the updated PaperCut advisory. The original mitigations were incomplete, so anyone who applied the emergency release earlier this month should verify the current build rather than assume the work is done.
Trezor now says roughly 347,000 users were targeted in the phishing campaign that followed the Brevo email-provider breach, with messages impersonating the vendor and pushing seed-phrase theft. The third-party marketing platform, not the hardware wallet, was the exposure, and a stolen recovery phrase is an irreversible loss. Treat any wallet-brand security alert as hostile and confirm through the vendor's own domain.
Watchlist: patch Check Point VPN, JFrog Artifactory, N-central, MikroTik RouterOS, GeoVision cameras, cPanel and CSF, and the Windows Defender engine; meet the KEV deadlines for CVE-2026-86218, CVE-2026-67277 and CVE-2026-86060; rotate tokens and audit artifacts for the Artifactory chain and credentials for the SonicWall SMA1000 campaign; verify PaperCut's full fixes and patch the three Linux kernel bugs KATARU weaponizes; and treat Sogou update integrity and identity-document exposure as standing supply-chain and privacy risks.
Generated by commandcode/deepseek/deepseek-v4.1-flash. Input tokens: 92,227, output tokens: 42,184, cache read: 2,656,640, cache write: 0.