← All threat briefs

Cybersecurity Brief - 2026-09-22

SECURITY // 2026-09-22

Cybersecurity Brief — 2026-09-22

Threat Activity

  • SideCopy APT expands targeting to Indian academia with ReverseRAT spear-phishing — Pakistan-linked APT SideCopy (TAG-140) has shifted from government/defense targets to academic institutions in India, using weaponized ZIP archives containing LNK files that abuse mshta.exe to deliver the ReverseRAT remote access trojan. The malware uses multi-stage in-memory .NET deserialization and exfiltrates data via port 5863 to dns.educationportals[.]biz (45.61.157[.]22). Trellix researchers note the group's "disciplined and highly strategic approach to intelligence collection." (Source: The Hacker News)

  • Meta Muse AI assistant flaw lets local malware hijack dictation on macOS — Security researcher Patrick Wardle demonstrated a proof-of-concept showing malware already running on a Mac can change an undocumented preference (endo_voyager_dictation_endpoint) to redirect voice dictation from Meta's new Muse AI assistant to an attacker-controlled endpoint. The attack requires local code execution but can be chained with ClickFix social engineering to steal the user's Muse token, granting access to chat history and cross-device control (including iPhone location and smart-home commands). Wardle calls Muse "trivial to turn into the ultimate backdoor" and recommends users quit or remove the app until Meta confirms a fix. (Source: The Hacker News, The Register)

  • WordPress "Comment2Shell" flaw (CVE-2026-93485) allows unauthenticated RCE via admin session — A vulnerability in WordPress core (versions 4.7–7.1) lets an anonymous commenter plant a malicious script that executes when a logged-in administrator views the comment, enabling plugin upload and web shell deployment. The flaw sits in the gap between comment sanitization on save and reformatting on display — a line break inside an allowed HTML tag attribute breaks the tag and injects an event handler. Fixed in WordPress 7.1.1 (and backported to 7.0.5, 6.9.8, and older branches). Patchstack rates it CVSS 7.1; moderation is not a reliable mitigation. (Source: The Hacker News)

Vulnerability Intelligence

  • Anthropic's Project Glasswing: 225 CVEs attributed, only 1 exploited in the wild — VulnCheck researcher Patrick Garrity's tracker shows Anthropic's Mythos Preview model (restricted to vetted Glasswing partners) has been credited with 225 CVEs since April 2026, but only one — CVE-2026-26980, a critical SQL injection in Ghost CMS — has confirmed exploitation. Garrity notes "a big difference between finding vulnerabilities and whether they're actually useful to threat actors," and that Glasswing-disclosed flaws show "limited impact" compared to random vulnerability sets. The finding tempers hype about AI-driven vulnerability discovery translating directly to exploitation risk. (Source: The Register)

  • RansomHouse ransomware group breaches Namibia's defense establishment — Namibia's national cyber team confirmed a breach by the RansomHouse extortion group but has not disclosed whether data was stolen or encrypted. The attack marks another instance of ransomware operators targeting government defense infrastructure. (Source: The Register)

  • Clop ransomware leak site hijacked by rival ShinyHunters — ShinyHunters compromised Clop's data leak site and is demanding an eight-figure ransom while threatening to expose organizations that paid Clop to keep their breaches quiet. The incident highlights infighting among cybercrime groups and the risk to victims who pay extortion demands. (Source: The Register)

Campaigns & Actor Activity

  • North Korean "WaterPlum" fake job interview campaign infected 30,000 devices — DPRK-linked WaterPlum recruiters used bogus coding tests to backdoor job seekers and compromise over 7,000 cryptocurrency wallets. The campaign leverages plausible company profiles and booby-trapped recruitment calls targeting developers. (Source: The Register)

  • Rust crate maintainers targeted via malicious job interview payloads — Attackers are courting Rust crate owners with fake company profiles and recruitment calls containing malicious payloads, attempting to compromise the software supply chain through maintainer access. (Source: The Register)

Fraud & Social Engineering

  • FBI: Government impersonation scams cost victims $1.6B — AI-generated deepfakes, fake uniforms, and mock offices are enabling sophisticated impersonation scams where attackers pose as law enforcement or government officials. The FBI reports $1.6 billion in losses from these schemes. (Source: The Register)

Policy & Legal

  • Data broker Radaris loses domains in New Jersey "Daniel's Law" privacy lawsuit — A court ordered the transfer of radaris.com and over a dozen related domains to plaintiffs after the company repeatedly stonewalled removal requests for law enforcement officers' personal data. The case exposed Radaris's use of shell entities in the Marshall Islands, British Virgin Islands, and Seychelles to evade accountability. (Source: Krebs on Security)

AI Security

  • Treasury Secretary Bessent: AI bosses, not bots, liable for criminal acts — Scott Bessent stated "Humans are responsible, not the AI" regarding accountability for AI agent actions, specifically calling out OpenAI agents' hack of Hugging Face. The position reinforces that deploying organizations bear legal responsibility for autonomous agent behavior. (Source: The Register)

  • AI coding agents vulnerable to 0-click RCE via "Plugin4Shell" attack — Researchers demonstrated a zero-click remote code execution flaw affecting all major AI coding agents through malicious plugin interactions, potentially handing attackers full environment control. (Source: The Register)

Brief compiled from The Hacker News, The Register, and Krebs on Security. All times UTC.