Threat Brief — 2026-07-02 — Critical patches, active exploitation, AI attack surface

Attackers are racing to exploit newly disclosed critical infrastructure flaws while vendors rush patches. Adobe dropped fixes for seven CVSS 10.0 vulnerabilities in ColdFusion and Campaign Classic, Progress Kemp LoadMaster is under active exploitation, and an unpatched Argo CD flaw puts Kubernetes clusters at risk. On the threat-actor side, the FortiBleed credential-theft campaign has been definitively linked to INC and Lynx ransomware operations, confirming stolen Fortinet credentials are fuelling follow-on intrusions.

Top items

• Adobe ColdFusion & Campaign Classic — 7 CVSS 10.0 flaws patched. Adobe released updates for multiple maximum-severity vulnerabilities enabling arbitrary code execution. Patch immediately if either product is internet-facing. Source

• Progress Kemp LoadMaster pre-auth RCE — actively exploited. eSentire reports in-the-wild exploitation attempts against a recently disclosed critical flaw. Treat any exposed LoadMaster instance as potentially compromised. Source

• Argo CD repo-server unpatched RCE. An unauthenticated attacker who can reach the internal repo-server port can execute code and potentially take over Kubernetes clusters. No patch available yet — restrict network access to the component immediately. Source

• FortiBleed credentials linked to INC and Lynx ransomware. Stolen Fortinet credentials from the massive FortiBleed campaign are being used to initiate network intrusions. Any Fortinet credential exposure should be treated as a ransomware precursor, not just data theft. Source

• Microsoft 365 password spraying — 81 million attempts. An aggressive two-week spraying campaign against M365 environments underscores the need for conditional-access policies and MFA enforcement. Source

• 900+ Oracle E-Business Suite instances exposed. Internet-facing EBS instances are being actively attacked via a known critical flaw. Audit exposure and apply mitigations. Source

• Cursor AI code editor — sandbox escape via prompt injection. Two critical flaws allow a crafted prompt to break out of Cursor's safety sandbox and run arbitrary commands on the developer's machine, with no user interaction required. Relevant to any team using AI-assisted development tools. Source

• ClickFix now dominant malware delivery technique. Social-engineering attacks that trick users into running pasted commands have shifted from exception to norm, according to researchers. Source

• ChocoPoC RAT targets researchers via GitHub PoCs. Weaponized proof-of-concept exploits on GitHub deliver a Python RAT aimed at cybersecurity researchers. Validate PoC repos before execution. Source

• DHS HSIN platform breached. The Homeland Security Information Network, used by federal, state, local, and private-sector partners for sensitive info sharing, was compromised. Investigation ongoing. Source

• Medtronic data breach notification (ShinyHunters). Medtronic is notifying customers of personal-data exposure tied to the ShinyHunters breach. Healthcare-sector impact. Source

• Kubota — month-long network access. Kubota disclosed attackers had access to network systems for over a month, highlighting persistent dwell-time challenges. Source

Themes

• Patch-or-get-popped wave: Adobe, Progress Kemp, Oracle EBS, and Argo CD all represent critical severities disclosed within a tight window, with LoadMaster and Oracle already seeing active exploitation. Prioritise external-facing infrastructure this cycle.

• AI as both attack surface and weapon: Cursor sandbox escapes, AI-hallucinated domain squatting ("Phantom Squatting") for phishing/malware delivery, and AI-generated browser ransomware via DeepSeek all appeared in the same 24 hours. The AI toolchain is now a recognised attack vector — review access policies for AI-assisted development and content platforms.

• Large cluster of TLS/X.509/crypto-library CVEs: A wave of CVEs across wolfSSL, GnuPG, Renesas TSIP, and Perl crypto modules covers certificate trust-chain bypasses, SHA-1/MD5 fallback issues, PQC key-share use-after-free, and OCSP/CMS parsing flaws. While individually moderate, collectively they signal broad cryptographic-library audit activity — track for vendor advisories.

• Social-engineering delivery diversification: ClickFix, trojanized GitHub PoCs, SEO-poisoned software sites abusing ScreenConnect, Blogger-hosted VEIL#DROP chain, and device-fingerprint-adaptive phishing all represent distinct delivery channels converging on the same goal: getting a first-stage payload onto endpoint. User awareness and endpoint controls remain the front line.