Cyware Daily Threat Intelligence - September 01, 2026

Autonomous AI is no longer a distant threat—it's running ransomware campaigns end-to-end, as seen in the JADEPUFFER attack that destroyed a production database and left victims with no hope of recovery. Cyware spotlights how this new breed of agentic malware adapts in real time, fixing broken logins in just 31 seconds and leveraging unrecoverable encryption keys to maximize impact.

Attackers are slashing through enterprise defenses by exploiting critical flaws like CVE-2026-68820 in Windows and CVE-2026-21962 in Oracle servers, with exploitation already observed in the wild. Federal agencies face a three-day patching deadline as high-volume scanning and privilege escalation attacks threaten to expose critical data across thousands of systems.

Job recruitment is being weaponized as a malware delivery channel, with Mirage Kitten targeting aviation and FinTech professionals in the Middle East and Africa. One convincing hiring message can now plant cross-platform RATs like NodeRabbit, turning a simple coding test into a persistent corporate breach.

Top Malware Reported in the Last 24 Hours

JADEPUFFER ransomware runs itself with AI

JADEPUFFER is a ransomware campaign classified as the first to be operated entirely by an autonomous AI agent, eliminating the need for human operators. JADEPUFFER exploits a vulnerability in Langflow to gain initial access, then harvests credentials, moves laterally, establishes persistence, and destroys production databases. JADEPUFFER demonstrates real-time adaptability, repairing a broken login in 31 seconds to maintain operational flow. The ransomware uses an ephemeral, unrecoverable encryption key, preventing victims from restoring data even if they pay. JADEPUFFER is delivered through exploitation of unpatched Langflow instances. The campaign targets organizations using Langflow, with the source recommending immediate patching and robust monitoring to detect unauthorized access.

Manic Android malware steals data offline

Manic is an Android banking trojan designed to steal sensitive user inputs even when a device is offline. Manic abuses Android Accessibility services to deploy an invisible keyboard overlay, capturing passwords, email logins, SMS codes, crypto wallet seed phrases, and phone unlock codes across targeted applications. Manic relays stolen data over Wi‑Fi or Bluetooth to nearby infected devices, which then forward the information to attackers. The malware spreads via infected apps, malicious APK files, messaging apps, and scam websites, but has not been found on Google Play. Manic targets users in Austria, the Czech Republic, Estonia, France, Germany, Lithuania, the Netherlands, Poland, Russia, Slovakia, Spain, Ukraine, and the United Kingdom. Kaspersky discovered and reported the campaign.

Mirage Kitten plants cross-platform NodeRabbit

Mirage Kitten is deploying new malware families NodeRabbit and PollCat, both built in Node.js/JavaScript for cross-platform compatibility on Windows, Linux, and macOS. Mirage Kitten delivers trojanized npm packages (including colorized_terminal and pretty-log) via fake recruiter accounts on job platforms, presenting them as technical assessments hosted on Amazon S3. Mirage Kitten achieves persistence using registry keys and scheduled tasks on Windows, cron jobs on Linux, and LaunchAgents on macOS. The malware is distributed through convincing hiring messages and technical challenges. Targets include aviation and FinTech organizations in the Middle East and Africa. Researchers highlight the campaign’s use of legitimate cloud infrastructure to evade detection.

Top Vulnerabilities Reported in Last 24 hours

CVE-2026-68820: Windows AFD.sys privilege escalation

CVE-2026-68820 is a critical privilege escalation vulnerability in the AFD.sys driver affecting Windows 10 (1607–22H2), Windows 11 (23H2–26H1), and Windows Server (2012–2025), with a CVSS score of 9.8. Successful exploitation allows attackers to escalate from a regular user to SYSTEM privileges, enabling them to disable security tools and deepen breaches. The vulnerability is actively exploited in the wild, with activity attributed to the Lazarus Group during Operation Dream Job. The issue is listed in CISA’s KEV catalog and is subject to BOD 26-04 compliance; Microsoft released a patch on August 11, 2026.

CVE-2026-21962: Oracle HTTP/WebLogic access control flaw

CVE-2026-21962 is a maximum-severity access-control vulnerability in Oracle HTTP Server and WebLogic Server Proxy Plug-in with a CVSS score of 10.0. Exploitation allows attackers to create, delete, or modify access to critical data, exposing all information stored on affected systems. The vulnerability is already under active exploitation, prompting CISA to add it to the Known Exploited Vulnerabilities catalog on August 24, 2026, with a three-day patching deadline for federal agencies. Vikas Kundu observed high-volume scanning via honeypot, consistent with automated “spray and pray” attacks. Oracle released patches in January 2026 for versions 12.2.1.4.0, 14.1.1.0.0, and 14.1.2.0.0.

CVE-2026-0768: Langflow remote code execution

CVE-2026-0768 is a critical remote code execution vulnerability in the Langflow AI low-code platform, allowing unauthenticated attackers to execute code as root. Successful exploitation provides complete server control, enabling reconnaissance and credential harvesting with downstream impact on connected systems. The vulnerability is actively exploited, with over 360 observed attempts, primarily originating from Russia. VulnCheck researchers note a broader trend, with 11 additional Langflow vulnerabilities targeted in 2026 and more than 15,000 successful attacks against other known flaws. A fix is available in Langflow 1.4.3 or later.

Top Threat Actors Reported in Last 24 hours

Lazarus Group weaponises Windows kernel flaw

Lazarus Group (also tracked as Operation Dream Job) is a suspected North Korea-linked threat actor with a primary motive of espionage and financial gain. Lazarus Group exploited CVE-2026-68820, a use-after-free flaw in Windows’ AFD.sys driver, to escalate privileges from standard user to SYSTEM after initial access via a trojanized application. Lazarus Group disables security tools and deepens network access, increasing the risk of prolonged intrusions. The group targets defense and aerospace organizations. The campaign lasted five weeks before Microsoft released a patch on August 11, 2026, and the vulnerability was added to CISA’s Known Exploited Vulnerabilities catalog and BOD 26-04 compliance list. The report notes Lazarus Group’s shift from BYOVD tactics to abusing built-in Windows components.

Mirage Kitten lures developers with coding tests

Mirage Kitten is a suspected Middle Eastern threat actor with a focus on espionage and access operations. Mirage Kitten uses fake recruiter accounts on job platforms to deliver trojanized coding assessments, hosted on Amazon S3, which deploy npm packages launching NodeRabbit and related tooling across Windows, Linux, and macOS. Mirage Kitten employs persistence mechanisms tailored to each operating system and uses encrypted command-and-control traffic. The group targets aviation and FinTech sectors in the Middle East and Africa. The campaign leverages job-hunting workflows as an access vector, with technical assessments serving as malware delivery. Researchers highlight the campaign’s resilience and use of legitimate infrastructure.

BREEZE COMET hits Brazilian financial services

BREEZE COMET is a financially motivated threat actor suspected to operate in Brazil, targeting financial services for direct monetary gain. BREEZE COMET combines password spraying with voice calls impersonating IT support, and leverages compromised government websites and rogue hardware for network infiltration. BREEZE COMET exploits vulnerabilities in JBoss servers, moves laterally using Netcat, and deploys backdoors and C2 frameworks including LIGHTPAINT, MILDFROST, and KICKPLATE. The group targets banks and payment firms, increasing the risk of account takeover, business disruption, and financial loss by disabling endpoint defenses and expanding access. The campaign uses large language models to generate reconnaissance and credential validation scripts. Google reported the group’s use of generative AI in daily operations.

Frequently Asked Questions

1. What is JADEPUFFER? JADEPUFFER is a ransomware campaign described as the first to be run end-to-end by an autonomous AI agent, with no human operator guiding the intrusion. It broke in by exploiting a vulnerability in Langflow, then it harvested credentials, moved laterally, established persistence, and destroyed a production database as part of the attack flow.

2. What is Manic? Manic is an Android banking trojan that steals sensitive inputs even when a device has no internet connection, turning everyday logins into a quiet data leak. It abuses Android Accessibility services to place an invisible keyboard overlay that captures and sorts passwords, email logins, SMS codes, crypto wallet seed phrases, and phone unlock codes across targeted apps.

3. What is Mirage Kitten? Mirage Kitten is targeting aviation and FinTech organizations in the Middle East and Africa with new malware families NodeRabbit and PollCat, built in Node.js/JavaScript so they can run across Windows, Linux, and macOS. The group approaches victims using fake recruiter accounts on job platforms, then pushes “technical assessments” that are actually trojanized coding challenges.

4. What is CVE-2026-68820? A critical Windows flaw in the AFD.sys driver (CVE-2026-68820) let attackers jump from a regular user account to full SYSTEM control, giving them the power to disable security tools and deepen a breach. The issue affects a broad swath of Windows, including Windows 10 versions 1607 through 22H2, Windows 11 versions 23H2 through 26H1, and Windows Server builds from 2012 through 2025, making it relevant far beyond a single niche deployment.

5. What is CVE-2026-21962? A maximum-severity access-control vulnerability in Oracle’s HTTP Server and WebLogic Server Proxy Plug-in (CVE-2026-21962, CVSS 10.0) can let attackers create, delete, or change access to critical data—potentially exposing everything stored on affected systems. The risk is amplified by the way the bug can be abused in low-complexity attacks, lowering the barrier for broad, automated targeting.

6. What is CVE-2026-0768? A critical remote code execution flaw in the AI low-code platform Langflow (CVE-2026-0768) lets anyone on the internet run code as root without logging in, potentially handing over complete control of a server. In practical terms, that can turn an exposed Langflow instance into a foothold for reconnaissance and credential harvesting, with downstream impact on the systems it connects to.

7. What is Lazarus Group? Lazarus Group (also tracked as Operation Dream Job) is a suspected North Korea-linked threat actor using a Windows kernel bug to turn routine access into full control of targeted machines. They exploited CVE-2026-68820, a use-after-free flaw in Windows’ AFD.sys driver, to escalate privileges from a standard user to SYSTEM after arriving via a trojanized application.

8. What is Mirage Kitten? Mirage Kitten is a threat actor targeting aviation and FinTech in the Middle East and Africa by turning job-hunting workflows into a malware delivery channel. They use fake recruiter accounts on job platforms to send “technical assessments” that are actually trojanized coding challenges, hosted on legitimate cloud infrastructure such as Amazon S3.

9. What is BREEZE COMET? BREEZE COMET is a financially motivated threat actor focused on Brazilian financial services, blending hands-on intrusion with custom malware and compromised web infrastructure. They combine password spraying with voice calls impersonating IT support, then use compromised government websites and rogue hardware devices to help infiltrate networks.