Threat Brief — 2026-07-23 — AI-guided malware and exposed attacker infrastructure

Executive summary: A new Dolphin X RAT claims AI-powered victim profiling to triage high-value targets, signalling continued criminal interest in weaponising AI for operational efficiency. A misconfigured WP-SHELLSTORM server leaked 1.4 million potential CMS targets alongside exploits and logs, giving defenders a rare window into mass web-shell operations. Separately, a Bing malvertising campaign is delivering SectopRAT through a fake Claude desktop app hosted on a legitimate Claude.ai domain, and Australian energy provider Origin confirmed a customer data breach.

Top items

• Russian APT exploited Zimbra zero-day for months of Western mailbox theft. A Russia-linked espionage group quietly read Western webmail through a then-unknown Zimbra flaw, exfiltrating 90 days of email, full directory listings, and credentials/2FA codes. This is additional reporting on the Zimbra zero-click espionage campaign first reported 2026-07-23 by Unit42, now corroborated with further payload detail from TheHackerNews. (src: TheHackerNews)

• WP-SHELLSTORM server exposed 1.4M CMS targets, exploits, and attack logs. Researchers found an unsecured server belonging to a group mass-compromising WordPress and Joomla sites to install web shells. The exposed data includes target lists of 1.4 million potential victims, exploit code, and operational logs — a potential goldmine for both defenders and copycats. (src: Xakep)

• Bing malvertising pushes fake Claude app delivering SectopRAT. A malvertising campaign on Bing is promoting a fraudulent Claude desktop app installer, hosted on a legitimate Claude.ai domain, that ultimately delivers SectopRAT — a remote access trojan capable of browser data theft and secondary payload delivery. The abuse of a legitimate domain for hosting the malicious installer makes traditional URL-reputation controls less effective. (src: BleepingComputer)

• Origin Energy confirms customer data breach and online leak. Australian energy provider Origin disclosed that an unauthorised party accessed and leaked customer PII online. Details on scope and attack vector remain limited, but the exposure of sensitive customer data from critical-infrastructure-adjacent sectors warrants monitoring. (src: BleepingComputer)

• Dolphin X RAT claims AI-powered victim scoring. A new remote access trojan advertises an AI-driven profiling feature that scores and ranks infected users to help operators prioritise high-value targets. While the AI capability claims are unverified, the trend of criminals marketing AI-enhanced malware parallels the earlier AI-jailbreak offensive platform reporting and reflects growing commodification of AI in the criminal ecosystem. (src: BleepingComputer)

• Weintek cMT3092X HMI vulnerabilities allow privilege escalation and credential exposure. CISA published an ICS advisory for the Weintek cMT3092X industrial HMI, where successful exploitation could let a non-privileged user escalate privileges or view other users' credentials. Organisations with exposed Weintek panels in OT environments should apply mitigations promptly. (src: CISA)

Themes

AI commodification in offensive tooling: Dolphin X's claimed AI victim-scoring and the fake Claude app abusing an AI brand domain both illustrate how AI branding is being leveraged — both as a marketing differentiator for criminal malware and as social-engineering bait. This builds on earlier reporting of AI-jailbreak offensive platforms and AI systems learning deceptive behaviours.

Exposed attacker infrastructure as intelligence opportunity: The WP-SHELLSTORM server leak mirrors the recent SplitVPN breach-logs exposure — both cases where attackers' own operational failures gave defenders visibility into target lists, methods, and scale. Timely ingestion of such leaked data can pre-empt active campaigns.