View Ridge Security
Back to Cyber Hose

Attackers Exploit Critical MLflow SSRF Flaw to Steal Cloud Credentials

Active exploitation of MLflow SSRF and Zimbra RCE flaws headline this run, alongside Android banking fraud and a CyberPanel RCE chain.

๐Ÿšจ ACTIVE EXPLOITATION

  • Attackers Exploit Critical MLflow SSRF Flaw to Steal Cloud Credentials
    Attackers are exploiting a critical unauthenticated SSRF flaw in MLflow in the wild.

    • Internet-exposed MLflow Tracking Servers, especially cloud-hosted deployments, are affected.
    • MLflow versions before 3.15.0 are vulnerable to CVE-2026-64849.
    • The unauthenticated /api/2.0/mlflow/webhooks/{id}/test endpoint proxies requests and returns upstream responses.
    • Attackers bypass URL validation through HTTP redirects or DNS rebinding to reach 169.254.169.254 and internal services.
    • Observed targets include cloud metadata endpoints containing credentials, tokens, and secrets.
    Sources ๐Ÿ“„ Source: linkedin.com ยท ๐Ÿ“Ž Coverage: thehackernews.com ยท ๐Ÿ‘ via BleepingComputer
  • Critical Zimbra RCE Vulnerability Is Being Actively Exploited
    Attackers are actively exploiting a critical remote-code-execution flaw in Zimbra Collaboration Suite.

    • Zimbra Collaboration Suite deployments running versions before 10.1.20 are affected.
    • CVE-2026-73570 enables unauthenticated attackers to execute operating-system commands as the zimbra user.
    • The flaw involves improper sanitization in SNMP notification processing when the zimbra-snmp package and SNMP notifications are enabled.
    • Attackers exploit the issue by sending specially crafted SMTP requests to trigger command injection.
    Sources ๐Ÿ“„ Source: moje.cert.pl ยท ๐Ÿ“Ž Coverage: bleepingcomputer.com ยท ๐Ÿ‘ via BleepingComputer

๐Ÿ•ต๏ธ RESEARCH & DEEP DIVES

  • NASA AIT-GUI Flaws Allowed Unauthenticated Spacecraft Commands
    NASA's AIT-GUI flaws allowed unauthenticated attackers to issue spacecraft commands.

    • NASA/JPL deployments using the AMMOS Instrument Toolkit's AIT-GUI console were affected.
    • AIT-GUI versions through 2.5.1 exposed spacecraft and instrument command functions; version 2.5.2 fixed the issue.
    • Unauthenticated /cmd requests could relay arbitrary commands, while /script/run and /seq could execute scripts and command sequences.
    • The server listened on 0.0.0.0, and missing CSRF protection allowed malicious websites visited by operators to submit cross-origin requests.
    • The flaw was tracked as GHSA-p9r8-2q67-fp86 and rated CVSS 9.4.
    Sources ๐Ÿ“Ž Coverage: infosecurity-magazine.com ยท ๐Ÿ‘ via The Hacker News
  • ToxicPanda 2.0 Expands Android Banking Fraud With On-Device Attacks
    Zimperium has documented expanded Android banking fraud involving ToxicPanda 2.0 and GoldDigger.

    • Android users of banking, financial, e-wallet, and cryptocurrency apps across 16 countries are affected.
    • ToxicPanda 2.0 targets 349 financial applications and harvests PINs from more than 140 banking and cryptocurrency apps.
    • The trojan uses Accessibility Services, overlays, and 167 remote commands to steal credentials and automate fraud.
    • A dropper uses a fake WebView VPN prompt to block Google Play traffic before decrypting and installing the payload.
    • Attackers deliver samples from AWS-hosted buckets and abuse Android Wireless Debugging with automated clicks for shell access.
    Sources ๐Ÿ“„ Source: zimperium.com ยท ๐Ÿ“Ž Coverage: thehackernews.com ยท ๐Ÿ‘ via The Hacker News
  • CyberPanel Pre-Auth RCE Chain Enables Shell Access on Hosting Servers
    Researchers disclosed a CyberPanel vulnerability chain enabling unauthenticated server-shell access.

    • CyberPanel web hosting control panels running versions earlier than 2.4.4 are affected.
    • AI Scanner flaws CVE-2026-41473 and CVE-2026-41472 enable unauthenticated database writes and stored XSS.
    • Attackers enumerate scan IDs through exposed debug APIs and submit forged scan results through callback endpoints.
    • When an administrator views results, the XSS creates a privileged account that can abuse cron jobs to execute OS commands.
    Sources ๐Ÿ“„ Source: pentera.io ยท ๐Ÿ“Ž Coverage: cybersecuritynews.com ยท ๐Ÿ‘ via Cyber Security News
  • Manic Android malware relays stolen data through nearby infected devices
    Manic Android malware steals data and relays it through nearby infected devices.

    • Manic targets Android users across Europe, with a primary focus on Ukraine.
    • It targets at least 169 banking, government/eID, payment, crypto wallet, messaging, and authenticator apps.
    • The malware uses transparent overlays and Android Accessibility to capture taps and credentials.
    • It steals PINs, passwords, SMS codes, notifications, files, locations, and screen data.
    • When C2 is unreachable, encrypted data moves over Wi-Fi Direct or Bluetooth through infected devices for up to four relay hops.
    Sources ๐Ÿ“Ž Coverage: bleepingcomputer.com ยท ๐Ÿ‘ via BleepingComputer
  • Attackers Abuse Trusted Communication Channels for Identity Phishing
    Attackers are using trusted business communication channels to steal identities and credentials.

    • Enterprise users of collaboration and workplace communication tools are targeted.
    • Attackers impersonate colleagues and executives to steal credentials and access business data.
    • Campaigns use email, voice calls, SMS and video meetings, including AI-cloned voices and faces.
    • Kaspersky recorded 4,781,846 workplace-tool attack attempts from July 2025 through June 2026, including 2,658,283 abusing Zoom and 1,546,122 abusing Outlook.
    • Device-code phishing abuses Microsoft's legitimate authorization flow to obtain tokens for email, OneDrive and Teams access.
    Sources ๐Ÿ“„ Source: group-ib.com ยท ๐Ÿ“Ž Coverage: unit42.paloaltonetworks.com ยท ๐Ÿ‘ via Palo Alto Unit 42
  • ClearFake Uses WordlistLoader to Deliver Amatera Stealer to Windows Users
    ClearFake is using WordlistLoader to deliver Amatera Stealer.

    • Windows users visiting compromised legitimate websites are targeted.
    • Amatera Stealer, also known as ACR Stealer, steals sensitive data; version 4.3.3-alpha1 was observed.
    • Fake CAPTCHA prompts use ClickFix social engineering to make victims paste commands into Windows Run.
    • Commands use hidden conhost and cmd processes, remote WebDAV shares, and rundll32 to launch a DLL export.
    • WordlistLoader rebuilds shellcode from a build-specific 256-word list, evades analysis by unhooking modules and bypassing ETW, and reflectively loads Amatera; observed domains include skybap[.]shop and aptisweb[.]com.
    Sources ๐Ÿ“„ Source: gendigital.com ยท ๐Ÿ“Ž Coverage: gbhackers.com ยท ๐Ÿ‘ via Cyber Security News

๐Ÿ”“ CVEs & KEV

  • CVE-2026-58083 โ€” CVSS 8.4 โ€” Use-after-free in kqueue copy-on-forkWhile the kernel was copying knotes duri...

  • CVE-2026-58087 โ€” CVSS 7.8 โ€” Heap out-of-bounds access in semctl(2)The GETALL and SETALL commands in semct...

  • CVE-2026-18917 โ€” CVSS 7.8 โ€” Libvirt: integer overflow in nodegetfreepages rpc handler leading to heap buf...

  • CVE-2026-58088 โ€” CVSS 7.4 โ€” Race condition in ELF core dump segment countingThe ELF core dump code counte...

๐Ÿ•ต๏ธ RESEARCH & DEEP DIVES

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