JFrog disclosed CVE-2026-105192, a critical flaw rated 9.8 in LMCache, open-source software that speeds up LLM servers such as vLLM. In multiprocess mode the cache runs as a standalone server reached over ZeroMQ with no authentication, and one message type is deserialized with pickle before its type is checked, letting a crafted message run code. The server is exposed only when an operator binds it to a routable address rather than the default localhost, and the code runs with LMCache's privileges, which are root in the project's official container images. No fix exists yet, and JFrog advises keeping the port on localhost or a trusted cluster network.
The FBI and Secret Service warned that FortiBleed, a credential-harvesting campaign against internet-facing FortiGate firewalls and SSL VPN gateways, remains active. The Russian-speaking operation uses credential stuffing and password spraying, then deploys a Go tool called FortigateSniffer to intercept authentication traffic, exploiting reused or leaked credentials and legacy SHA-256 password storage rather than a software flaw. It had collected more than 86,644 working device credentials across 194 countries as of June. Attackers create new admin accounts and reuse session cookies for persistence, and some victims are locked out when original accounts are changed or deleted. Operator overlaps link it to INC and Lynx ransomware.
Lumen Black Lotus Labs detailed PoeLLM, the malware behind the financially motivated Canto Incognito campaign, which has infected more than 3,400 internet-facing servers since April 2026. The actor hides its command-and-control address inside a poem stored in a GitHub repository, changing a few words whenever it rotates infrastructure. It exploits known vulnerabilities in exposed services such as LiteLLM, Gotenberg, Gitea, and Ivanti Sentry, then installs XMRig and Iron miners to abuse victim compute for cryptocurrency mining through a Russian mining service. Compromised hosts become scanners and exploit servers that spread the malware further, and recent traffic suggests experimentation with distributed brute-force attacks.
CloudSEK and Checkmarx detailed MALFEX, a campaign they attribute to a lone actor who has published eight malicious npm packages downloaded 40,767 times in total, with function-flag alone accounting for 37,419. The Windows-focused packages deliver three payloads: Overlord, an open-source Go remote access trojan that pulls its command-and-control address from Solana transactions; a Node.js stealer that targets Discord, browsers, Telegram, and cryptocurrency wallets; and a downloader. Lifecycle and postinstall hooks fetch and run remote executables, with one package saving a hidden node.exe to the AppData directory. Three packages, function-flag, function-color, and cdn-img-fetch, were still live on npm at publication.
Atlassian disclosed CVE-2026-21589, a path traversal flaw rated 9.3 that lets an attacker with no login access read specific files in each product's web application root directory when the exact path is known. It affects the Data Center editions of Jira Software, Jira Service Management, Confluence, Bitbucket, Bamboo, Crowd, Crucible, and Fisheye. After watchTowr published a proof of concept, security firm Previdian observed exploitation attempts against its honeypots within two hours. In Crowd-integrated Jira deployments, attackers extract plaintext credentials from configuration files and use Crowd's API to create administrator accounts, taking full control of the instance.
SonicWall released hotfixes for CVE-2026-102255, an unauthenticated server-side request forgery flaw in the Appliance WorkPlace interface of its SMA1000 secure access gateways. A remote attacker with no credentials could use an unintended alternate access path to direct the appliance to issue requests on their behalf, reaching internal functionality and performing unauthorized operations. The flaw affects SMA1000 6210, 7210, and 8200v models and does not impact the SMA 100 series or firewall SSL-VPN. SonicWall says there is no evidence of exploitation yet, but more than 400 internet-exposed SMA1000 appliances are tracked and the line has historically drawn ransomware operators.
Microsoft patched CVE-2026-96940, a high-severity flaw rated 8.8 in Exchange Server that lets an authenticated attacker gain unauthorized access to other users' mailboxes within the same organization. The weakness stems from weak authorization controls that permit privilege escalation, exposing colleagues' messages and attachments. Affected builds include Exchange Server Subscription Edition RTM, Exchange Server 2016 Cumulative Update 23, and Exchange Server 2019 Cumulative Updates 14 and 15. Microsoft already applied a related service-side fix to Exchange Online, so on-premises administrators must install the updates themselves. There is no evidence of active exploitation, but Microsoft rated the flaw Exploitation More Likely.
Patchstack identified a campaign on October 4 exploiting CVE-2026-94504, a high-severity stored cross-site scripting flaw in the Ninja Forms WordPress plugin, which is installed on more than 500,000 sites. Attackers plant malicious JavaScript in form submissions, and when a logged-in administrator loads the content, the script runs under the authenticated WordPress session. The payload installs a backdoor plugin and creates several hidden access paths: a visible administrator account, a concealed account kept out of the user list, a secret login URL, and unauthenticated file manager access. Versions 3.15.3 and older are affected, with a fix in 3.15.4.
Researchers disclosed CVE-2026-63277 in LibreOffice and CVE-2026-59265 in Apache OpenOffice, flaws that let a malicious spreadsheet run code without triggering macro warnings. The attack chains legitimate features: a spreadsheet defines a database range that downloads an ODB database file from a web address, and that file references a JDBC driver, causing automatic download and execution of attacker-controlled Java code. LibreOffice fixed the issue in versions 26.2.5 and 26.8.0 on October 5, while Apache OpenOffice remains affected through 4.1.16 with 4.1.17 still in testing. The technique exists only as a proof of concept, credited to researchers at V12 and Codean Labs.
Microsoft Threat Intelligence detailed a ClickFix variant that pre-fetches malicious scripts into the browser cache, disguised as PNG files, so victims run content already stored locally rather than downloading a remote payload when they paste the attacker's command. The approach conceals the payload and sidesteps the roughly 260-character limit of the Windows Run dialog. The chain uses VBScript to enumerate browser cache files, pulls PowerShell from external servers, and injects .NET assemblies into legitimate processes such as timeout.exe to steal browser and device credentials before fetching further in-memory stages. The cache enumeration specifically references Firefox, and similar tricks have targeted macOS Terminal users.