🔴 NSA/CISA warn of AI-powered PLC attacks, 🎯 Citrix NetScaler CVSS 9.3 auth bypass, ⚠️ Zimbra RCE actively exploited, 🛡️ CyberPanel pre-auth RCE chain, 🔓 Splunk MCP Server critical RCE, 📋 Clop Windchill web shell + long tail emerges

NSA, CISA, FBI, DOE, and EPA issued a joint advisory warning of active campaigns targeting Siemens S7 Series PLCs across U.S. critical infrastructure sectors including water, energy, chemical, and manufacturing. Attackers are using internet scanning services to find exposed controllers, then deploying AI-generated Python exploitation scripts to target critical and high-severity vulnerabilities, outdated firmware, and weak authentication. The advisory explicitly states the threat extends beyond Siemens to all PLC operators, and the use of LLM-assisted tooling dramatically lowers the technical barrier for ICS exploitation. Separately, NSA and CISA published a companion advisory on active threats targeting Siemens S7 controllers with specific MITRE ATT&CK techniques and IOCs for network detection. Both advisories emphasize that internet-exposed PLCs are particularly dangerous given the permissive-by-design communications protocols and limited built-in authentication. Organizations running Siemens S7 controllers should verify these devices are not directly internet-reachable, review IT/OT network segmentation, and check for anomalous Modbus/TCP or S7comm traffic patterns.

[REVISIT] Cloud Software Group disclosed Citrix NetScaler CVE-2026-19490, a CVSS 9.3 authentication bypass that allows unauthenticated attackers to access customer-managed NetScaler ADC and Gateway appliances without valid credentials. CVE-2026-19489 is a high-severity denial-of-service flaw affecting the same products. Given the history of NetScaler being targeted by ransomware operators and nation-state actors including the 2023 Citrix Bleed campaign, exploitation is expected once technical details become available. Cloud Software Group urges immediate patching of all customer-managed instances.

[REVISIT] CERT Polska issued Alert 145/2026 warning that Zimbra CVE-2026-73570 is being actively exploited in the wild. The flaw is an unauthenticated OS command injection in the SNMP monitoring component (swatchdog service) that allows attackers to execute arbitrary shell commands as the zimbra user. The Zimbra security team patched this in version 10.1.20 released July 20, but exploitation began before most organizations rolled out the update. Swatchdog is enabled by default in many deployments, meaning the vulnerable code path is active unless administrators explicitly disabled SNMP monitoring.

[REVISIT] Two independent research groups disclosed a pre-authentication RCE chain in CyberPanel chaining CVE-2026-41473 (authentication bypass) and CVE-2026-41472 (stored XSS) through the AI Scanner component and built-in cron-job functionality to achieve host-level command execution as root. CyberPanel operates as an administrative layer over web-hosted services, so a successful compromise means full server control including all hosted websites, databases, and email accounts. Organizations should disable or restrict access to the AI Scanner endpoint immediately.

Splunk released security hardening advisory SVD-2026-0808 patching 17 vulnerabilities, the most severe being CVE-2026-76404, a CVSS 9.1 RCE in the Splunk MCP Server app (versions before 1.2.1). The flaw allows a user with the Splunk admin role to execute arbitrary OS commands due to insufficient input validation in the credential-management component. Additional CVEs address privilege-management, authorization, deserialization, and data-exposure weaknesses in the Splunk AI Toolkit. Organizations should prioritize updating the MCP Server app to 1.2.1 or later and audit which of the 17 affected add-ons are installed.

[UPDATE] Clop ransomware operators built a purpose-built Java web shell tailored to PTC Windchill's internal APIs, database schema, and file-vault structure to exploit CVE-2026-12569 (CVSS 9.3), and Shell, General Electric, and Philips are now confirmed investigating the campaign. The implant is not a generic web shell repurposed for the campaign but was engineered with deep knowledge of Windchill's credential storage, file-vault enumeration, and data exfiltration paths, giving Clop a single tool that can decrypt stored credentials, enumerate engineering repositories, and steal files without additional attacker infrastructure. The targeting of PLM systems that store manufacturing intellectual property elevates this beyond a typical file-transfer exploitation campaign. The long tail of Clop's PTC hack is just beginning to emerge as affected organizations complete forensic investigations.

[UPDATE] GuidePoint Security's GRIT team uncovered Ransom Busters, a double-extortion tactic where a ransomware affiliate impersonates a third-party recovery service, contacting victims before their attacks become public, claiming to have breached ransomware infrastructure, recovered decryption keys, and deleted stolen backups. The affiliate targets victims of DragonForce, Settra, and Anubis families, siphoning payments away from the RaaS platform. GRIT assesses with moderate confidence that Ransom Busters is the affiliate itself.

A cloud-only business email compromise campaign targets finance employees using Microsoft 365 by exploiting Adversary-in-the-Middle phishing kits to capture authenticated MFA session tokens, then creating hidden inbox rules that auto-delete or redirect emails related to vendor payments. The attack leaves no malware on the endpoint and no endpoint compromise to detect, living entirely in the cloud through session token theft, inbox rule manipulation, and wire fraud redirection to attacker-controlled bank accounts.

Researchers identified Aeternum, a C++ botnet loader that abuses Polygon blockchain smart contracts as decentralized C2 infrastructure. Instead of contacting traditional attacker-owned servers, infected machines send JSON-RPC requests to public Polygon RPC endpoints and read contract values to retrieve C2 domains. This dead-drop resolver pattern makes takedowns significantly harder because there is no single domain to seize or server to sinkhole, and the contract data lives on a public, immutable ledger.

[REVISIT] TheGentlemen ransomware group has now claimed 10 victims across six countries since August 13, up from the original batch of seven. The three newest claims, all posted August 19, add Senvest Capital (international investment firm, USA), Roadvision Systems (trucking management software, USA), and Gfeller Treuhand und Verwaltungs AG (Swiss real estate and fiduciary services). The expansion from seven to ten victims in under a week points to a sustained operation targeting mid-sized companies in non-IT verticals including recycling, HR, food service, auto dealerships, and investment firms.

Hunt.io uncovered a campaign dubbed CameraSwarm that compromised 14,530 Dahua IP cameras over 35 days primarily in Ukraine and Russia. Three parallel attack methods drove the campaign: brute-forcing TCP port 37777 compromised 12,324 cameras, CVE-2021-33044 and CVE-2021-33045 exploitation installed a persistent backdoor account on 1,923 cameras that survives password changes and factory resets, and cloud-relay exploitation reached 283 cameras behind NAT. The persistent backdoor surviving factory resets is a serious remediation challenge requiring hardware replacement or firmware-level intervention.

Security researcher Eric Chiang used Anthropic's Claude Opus in an AI-assisted vulnerability research pipeline and found critical authentication bypass flaws across multiple open-source SAML implementations. The findings include full authentication bypasses, XML signature validation weaknesses, information disclosure, and denial-of-service conditions, all stemming from parsing differential vulnerabilities between signature-validation libraries and relying applications. SAML's fundamental weakness is its dependence on XML Digital Signatures, and mismatches in how applications parse or canonicalize XML documents can let attackers slip in attacker-controlled content that the application accepts as legitimate.

[REVISIT] ClearFake campaigns are using a new loader called WordlistLoader to deploy Amatera Stealer. The technique hides shellcode inside lists of ordinary English words, where each word maps to a single byte value based on its position, making static detection harder because the raw data looks like innocuous text. The ClearFake operation uses advanced anti-analysis mechanisms alongside the WordlistLoader obfuscation, marking a shift from earlier Python-based loaders to a more stealth-oriented approach.

[REVISIT] Wiz Red Agent, an autonomous AI security research tool, discovered a critical GitHub Actions injection vulnerability in Snowflake's public snowflake-connector-net repository. The flaw in jira_issue.yml allowed an unauthenticated attacker to execute arbitrary commands on the GitHub Actions runner and exfiltrate credentials for Snowflake's internal Jira environment. The workflow was deployed and the flaw discovered just five days later. Snowflake patched the vulnerability after Wiz's disclosure.

In 2024, T-Mobile's security team physically cut a network cable to sever access by Salt Typhoon, the Chinese state-sponsored espionage group targeting telecommunications providers worldwide. The physical disconnection was a last resort after conventional network-level controls could not fully contain the intrusion, underscoring the severity of Salt Typhoon's campaign against US telecoms and serving as a reference case for incident response at the boundary of network and physical security.

The DOJ unsealed a 14-count indictment charging 17 Iranian nationals linked to the Mabna Institute with conducting cyber theft campaigns targeting American and foreign institutions on behalf of Iran's Islamic Revolutionary Guard Corps. Five of the defendants have $10 million rewards posted for information leading to their arrest, signaling the FBI considers these individuals high-priority targets with firm attribution.

A new Android malware named Manic targets at least 169 banking, government, payment, crypto wallet, messaging, and authenticator apps, with users in Ukraine as the primary focus. The malware uses transparent overlays on numeric keypads to capture PINs and passwords, and includes a relay mechanism that exfiltrates data through nearby infected devices when direct internet access is blocked.

Red Hat disclosed CVE-2026-66794, a CVSS 9.3 server-side request forgery vulnerability in the cluster-proxy-addon of the Multicluster Engine for Kubernetes, allowing unauthenticated remote attackers to access otherwise isolated services across managed clusters through publicly accessible routes.

[UPDATE] Storm ransomware claimed a fifth victim in four days, hitting WindRose Health Network alongside Rood & Riddle Equine Hospital, Tapper Cuddy LLP, Integra Castings, and Westco Motors Cairns. The breadth across healthcare, legal, manufacturing, and automotive sectors in three countries simultaneously suggests either a broad initial access campaign or an affiliate model feeding multiple operators into the same leak site.

Watchlist: apply the VMware vCenter patches (CVE-2026-59310, CVE-2026-59309) before the August 21 CISA deadline given active exploitation with ransomware deployment; prioritize the Microsoft AFD.sys zero-day (CVE-2026-68820) which requires a reboot to fully remediate; patch BeyondTrust EPM to 26.1.2 before the privilege-escalation chain is weaponized; patch GitLab to address CVE-2026-19478 before exploitation scales; verify Zimbra instances are on version 10.1.20 or later; treat Citrix NetScaler CVE-2026-19490 as emergency maintenance; Siemens PLC operators should verify internet exposure and review IT/OT segmentation; and organizations running open-source SAML libraries should audit their XML canonicalization stacks.