Threat Brief — 2026-09-05 — Unpatched store zero-day, AI-assisted attacks evolve

Executive summary: An unpatched zero-day in Magento Open Source and Adobe Commerce is being actively exploited to backdoor online stores without authentication — the most urgent item today. Unit 42 has identified two additional threat groups using AI in real-time during attacks, building on earlier findings. Separately, X's new payment feature has transformed account hijacking from an audience-theft problem into a direct financial crime opportunity.

Top items

• Actively exploited unpatched Magento / Adobe Commerce zero-day. Attackers are leveraging a new, unpatched vulnerability in Magento Open Source and Adobe Commerce that allows unauthenticated remote code execution on store servers, enabling backdoor installation. Sansec reports active in-the-wild exploitation. No CVE or patch is available yet. This is the first time this story has been featured despite appearing in feeds earlier today. (src: The Hacker News)

• Unit 42 identifies two more groups using AI during live attacks. Building on its earlier disclosure of AI-assisted enterprise intrusion (first reported 2026-09-02 by Unit 42 Threat Research), Unit 42 now reports catching two additional groups deploying AI directly during active attack operations. The evidence indicates AI is accelerating attack execution, though attackers have not yet improved their operational security to match. (src: SecurityLab.ru)

• X Money turns account hijacking into direct financial theft. X's rollout of X Money means that hijacking a user's X account no longer merely yields audience access or impersonation capability — it can now provide a path to real funds. This shifts the economic calculus for credential-stealing campaigns targeting the platform. (src: SecurityLab.ru)

• Malicious ads bypass Google moderation at scale; AI used to catch them. Researchers ran 188,000 advertisements through Google's Gemma model and discovered thousands of malicious materials that had already passed human and automated moderation, collectively reaching approximately 100 million impressions before detection. This highlights both the scale of adversarial ad abuse and the potential for defensive AI screening. (src: SecurityLab.ru)

• Ukraine cyber police chief arrested for running scam call-center network. Ukraine's National Anti-Corruption Bureau detained the head of the cyber police in connection with the "Carthage" case, alleging he operated a network of fraudulent call centers. The case has now reached the Prosecutor General's Office. (src: SecurityLab.ru)

Themes

AI as both weapon and shield. Today's findings reinforce a dual trajectory: AI is being embedded into live attack operations by multiple groups (Unit 42's updated findings), while simultaneously proving useful for defensive screening at scale (the Gemma ad-detection work). Proposals for physically isolating dangerous AI models via network diodes also surfaced, reflecting growing concern that sandboxing alone is insufficient.

Monetisation pathways expanding. The X Money development illustrates how platform feature additions can inadvertently create new monetisation paths for existing attack techniques — account takeover is the same, but the payoff has changed.