CISA added CVE-2026-8037 to its Known Exploited Vulnerabilities catalog on August 7, 2026, and set a federal remediation deadline of August 10. Progress disclosed the vulnerability on June 4 alongside CVE-2026-33691. WatchTowr Labs published a technical analysis in June describing the root cause as an improper null-termination in a function named escape_quotes(), which fails to properly bound-check sanitized strings, causing an out-of-bounds read into adjacent heap memory. That read allows an attacker to issue specially crafted requests to the /accessv2 endpoint and achieve command injection without any credentials. eSentire's Threat Response Unit observed the first exploitation attempts on June 29, the day WatchTowr published its analysis. Those early attempts failed and produced no confirmed post-compromise activity. KEVIntel telemetry captured 792 exploitation attempts from 65 unique IP addresses across 18 countries between June 29 and August 4, with the last recorded activity on August 4. CISA's KEV record pairs the patch requirement with its forensics triage guidance, treating this as an incident response item rather than a routine patch event. LoadMaster is an application delivery controller positioned at the network edge in front of web applications, IIS deployments, Exchange environments, and other internal services. Compromise of the appliance gives an attacker a foothold with visibility into internal service topology and the potential for lateral movement into any service LoadMaster proxies. CVE-2026-8037 also affects ECS Connection Manager, Connection Manager for ObjectScale, and MOVEit WAF. Fixed versions are GA 7.2.63.2 and LTSF 7.2.54.18.

Between August 6 and August 7, 2026, a threat actor published nearly 800 packages to the npm registry within 48 hours. OpenSourceMalware researcher Paul McCarty disclosed the campaign and described the package names as AI-slop squatted, meaning they use AI-generated combinations that resemble plausible package names without targeting specific well-known ones, a technique that avoids triggering typosquat detection tools keyed to known package names. All packages deliver a downloader named WEL1DROPPER. Unlike most malicious npm packages that use preinstall or postinstall lifecycle hooks to run code automatically, WEL1DROPPER loads through the main module entrypoint. The README instructs developers to import the package with require(), and that single call starts the infection chain with no hook required. Upon execution, WEL1DROPPER identifies the host operating system and processor architecture and fetches the appropriate payload from one of three Cloudflare Workers hosts. If those HTTPS downloads fail, the malware switches to DNS TXT record delivery through subdomains of wel1.ru, reconstructing the payload from record contents to bypass network controls that block the primary delivery channels. The final payload is a cross-platform RAT and infostealer compatible with Windows, macOS, and Linux. On macOS the malware drops a fake runtime binary and a LaunchAgent for persistence. The beacon payload was not fully available for analysis at disclosure time, meaning the complete range of post-installation behavior remains uncharacterized.

Varonis Threat Labs disclosed RovoBlast at DEF CON and published the full write-up on August 5. Rovo is Atlassian's enterprise AI assistant, spanning Jira, Confluence, Bitbucket, and more than 50 third-party platforms through Rovo Connectors, including Slack, Microsoft 365, and Google Workspace. The vulnerability was a URL parameter named rovoChatPrompt that pre-filled content into Rovo Chat before a user opened a session. An attacker could craft a link placing a complete set of attacker instructions into that parameter. When an authenticated user clicked the link, Rovo treated the parameter contents as trusted input within that user's active session and executed the instructions with the user's credentials and permissions. Varonis' proof of concept showed Rovo enumerating accessible Jira tickets, Confluence pages, and SharePoint documents and exfiltrating their contents by embedding the data in requests to an attacker-controlled image URL, which Rovo fetched as part of its research workflow. The attack required no jailbreak, no bypass of Rovo's guardrails, and generated no warning to the user. Atlassian patched the rovoChatPrompt issue server-side on July 8, 2026, after disclosure through Bugcrowd. No customer action is required; all Atlassian Cloud tenants with Rovo enabled received the fix automatically. A separate content-borne prompt injection chain was reported to Atlassian by PromptArmor in May 2026 and disclosed publicly on August 5. That chain uses attacker instructions embedded in documents or pages Rovo processes, rather than a URL parameter. PromptArmor stated that chain was still unpatched at the time of publication. Atlassian's remediation status for the content-borne path has not been confirmed as of August 10.