Cyber Wargame Simulates Attack on US Water; Ransomware Accelerated by Agentic AI

Andy Greenberg has published his account of a CyberAcuView wargame simulating a cyberattack on US water infrastructure, conducted in April 2026 by Joshua Corman, a former strategist for the US Cybersecurity and Infrastructure Security Agency (CISA) and current Executive in Residence for Public Safety & Resilience at the Institute for Security and Technology (IST) think tank. The scenario imagines attacks taking place in July 2027, causing outages over three days impacting 5,000 water utilities, hobbling the supply chains for food and medicine, taking down cloud services, leaving 2,000 hospitals without water, and causing physical destruction like burst water mains. Greenberg compares Corman to a roleplaying game's "dungeon master," giving momentum and pressure to the simulation with curveball challenges, demands from non-player stakeholders, and even uncertainty, such as a dice roll that determines whether incident responders from major firms are available or not. The players were ~30 insurance executives, whose "surprisingly pivotal role" during the fallout from a disruptive cyberattack on water utilities is to communicate with their clients and allocate resources, including cyber incident responders, legal counsel, and money. The insurers discussed how to triage clients and responses as new factors arose affecting ethics, financial risk, and liability; priorities included public health and safety, business relationships and client revenue, pressure from regulators and shareholders, and military focus and "act of war" exclusions if the attack were an act of cyberwarfare. Greenberg also provides background on evidence of real-world threats to US critical infrastructure from PRC state-sponsored threat group Volt Typhoon, the implied instigator in Corman's simulation. Corman states that this scenario is not designed to be won or lost, but to unearth participants' assumptions, as while there are actions to take that make a significant difference, "if even a relatively mild version of Volt Typhoon’s potential hacking occurred under current conditions ... it would be beyond the insurance industry’s capacity to handle it." Corman also hopes the difficulty of response to catastrophe shifts some focus toward prevention, such as changes in insurance policies that compel clients to meet a higher standard of cybersecurity and to join information sharing groups.

In a July 1 report, Sysdig's Michael Clark writes that "the Sysdig Threat Research Team (TRT) has captured what [they] assess to be the first documented case of agentic ransomware: a complete extortion operation driven end-to-end by a large language model (LLM)." Dubbed JadePuffer, the operator exploited two known vulnerabilities — one in Langflow (CVE-2025-3248) and a second in Nacos (CVE-2021-29441) — to ultimately encrypt files on a production database server. In an interview with CyberScoop, Clark acknowledges that the attack included human involvement: "A human still set up and pointed the operation and provisioned the infrastructure behind it, the command-and-control server, the staging server used for the stolen data and chose a victim." The attack is notable for the speed and scope of the agent's activity, as well the fact that the payloads "contained natural language reasoning, target prioritization, and the kind of detailed annotations that human operators don’t often write but LLM-generated code produces reflexively." Sysdig's recommendations include patching and hardening systems against the known vulnerabilities, ensuring that database servers' administrative accounts are not exposed to the internet, applying egress controls, and using runtime threat detection.

The US Cybersecurity and Infrastructure Security Agency (CISA) added four CVEs to the Known Exploited Vulnerabilities (KEV) database on Tuesday, July 7. Federal Civilian Executive Branch (FCEB) agencies are expected to mitigate all four vulnerabilities by Friday, July 10. CVE-2026-48282 is a critical (CVSS score 10.0) path traversal flaw in Adobe ColdFusion that could be exploited to achieve arbitrary code execution in the context of the current user. The issue affects ColdFusion versions 2025.9, 2023.20 and earlier. Adobe fixed the issue in ColdFusion 2025 Update 10 and ColdFusion 2023 Update 21; the updates were released on June 30, 2026. CVE-2026-55255 is a high-severity (CVSS score 8.4) authorization bypass vulnerability in Langflow that could be exploited by an authenticated attacker to execute any flow belonging to another user by specifying the victim's flow ID in the request. The issue is fixed in Langflow 1.9.1. CVE-2026-56290 is a critical (CVSS score 9.8) improper access control vulnerability in the Joomla Page Builder CK extension that could be exploited to achieve remote code execution via unauthenticated arbitrary file upload. The vulnerability was addressed in Page Builder CK version 3.6.0, which was released on June 27, 2026. CVE-2026-48908 is a critical (CVSS score 9.8) improper access control vulnerability in the Joomla SP Page Builder extension that could be exploited by unauthenticated users to upload arbitrary files.

Carnegie Mellon University's CERT Coordination Center (CERT/CC) has published a Vulnerability Note warning that "Several versions of Tenda firmware contain an undocumented authentication backdoor that grants administrative access to the devices' web management interfaces. An attacker can exploit this vulnerability, tracked as CVE-2026-11405, to bypass the password verification process and obtain full administrative control without valid credentials." The issue, which is due to an undocumented authentication mechanism in the 'login()' function of the '/bin/httpd' web server binary, could be exploited to gain full administrative access to a vulnerable device's web interface "regardless of the configured administrator account credentials." CERT/CC writes that they have been unable to reach the Chinese manufacturer and that no patch is currently available. Their suggested mitigations include disabling remote management on vulnerable devices and restricting local network exposure.

Cyber threat actors with ties to China exploited critical vulnerabilities in the Roundcube open-source email client to gain access to university systems in Canada and the US. The attackers chained the two vulnerabilities — CVE-2024-42009, CVSS score 9.3, a cross-site scripting issue, and CVE-2025-49113, CVSS score 9.9, a deserialization of untrusted data issue — to execute JavaScript in targeted browsers and gain purchase in the mailserver. The attackers appear to have deliberately targeted institutions whose Roundcube systems had not been patched against the vulnerabilities, which suggests they conducted reconnaissance prior to the attacks. Proofpoint says the attackers are stealing credentials and "install[ing] a webshell for follow-on access or deploy[ing] the VShell backdoor into the server’s memory." The threat actors have designed the campaign to evade detection. The targets appear to be physics and engineering departments at the universities, with a focus on professors and administrators with ties to national security, and on institutions conducting research in astrophysics and particle physics. Researchers at Proofpoint first detected the campaign in May of this year, and they are calling the threat cluster “UNK_MassTraction.”

Researcher Hyunwoo Kim discovered and disclosed a flaw in the kernel-based virtual machine (KVM) module in the Linux kernel that has made systems vulnerable to VM escape since 2010. Dubbed "Januscape," CVE-2026-53359 allows a guest-side attacker "to corrupt the host kernel's shadow page, and ... threaten the guest-host isolation of KVM/x86 hosts that accept untrusted guests and expose nested virtualization, particularly multi-tenant x86 public clouds (GCP, AWS, etc.)," due to a use-after-free vulnerability in the shadow MMU emulation of KVM/x86. This flaw affects both Intel and AMD architectures, but not arm64. Kim has published a proof-of-concept (PoC) exploit for causing host kernel panic using guest actions alone, and says that a PoC for a full escape exploit exists but will only be released "in the distant future." Kim also notes that CVE-2026-53359 was successfully used in Google's kvmCTF vulnerability reward program; Google's reward offered for full VM escape vulnerabilities in this program is $250,000. Another flaw discovered separately by Nebula Security, also disclosed through kvmCTF, has been present in the kernel since 2011: CVE-2026-43499 ("GhostLock," CVSS 7.8) allows an unprivileged local attacker to achieve root access by exploiting a use-after-free vulnerability in rtmutex/futex-PI. Nebula states that their researchers discovered this flaw using their VEGA AI vulnerability scanning tool, and that Google paid a reward of $92,337. Users should check their distribution for updates, as both flaws have been patched in the Linux kernel.

BeyondTrust has released updates to address four security issues in its products, including a pair of critical vulnerabilities in its Remote Support (RS) and Privileged Remote Access (PRA) software. CVE-2026-40138, CVSS score 9.2, is an improper authentication issue in the authentication subsystem of RS and PRA that in a certain configuration could be exploited to bypass access controls and gain unauthorized access to a vulnerable appliance. CVE-2026-40139, CVSS score 9.2, is an improper authentication issue in the authentication subsystem of RS that in a certain configuration could allow a remote attacker to gain unauthorized access to a vulnerable appliance. CVE-2026-40140, CVSS score 8.7, is an uncontrolled resource consumption issue in RS and PRA that could be exploited to create a denial-of-service condition on a vulnerable appliance. CVE-2026-40141, CVSS score 8.5, is an improper neutralization of special elements in data query logic issue in a web application component of RS and PRA; this vulnerability is related to the way certain input parameters are processed, and could be exploited by an "authenticated attacker with limited privileges to access unintended resources or data beyond their authorization scope." BeyondTrust applied a patch for the vulnerabilities for RS and PRA cloud customers on April 21, 2026; self-hosted customers who are not subscribed to automatic updates are advised to ensure that the appropriate April rollup patch is applied or upgrade to RS 25.3.3 and above or PRA 25.3.3 and above.

Ubiquiti has released updates to address seven critical vulnerabilities in UniFi Connect, UniFi Talk, UniFi Access, UniFi Protect, and UniFi OS. CVE-2026-50746, CVSS score 10.0, is an improper access control issue in UniFi Connect Application that could be exploited to execute command injection. CVE-2026-50747, CVSS score 9.9, is a series of SQL injection issues in UniFi Talk Application that could be exploited to escalate privileges. CVE-2026-50748, CVSS score 9.9, is an improper input validation issue that could be exploited to execute command injection. CVE-2026-55115, CVSS score 9.9, is a server-side request forgery issue in UniFi Protect Application that could be exploited to escalate privileges. CVE-2026-54402, CVSS score 9.9, is an improper access control issue in UniFi OS that could be exploited to execute command injection. CVE-2026-54400, CVSS score 9.1, is an improper access control issue in UniFi Access Application that could be exploited to escalate privileges. CVE-2026-55116, CVSS score 9.0, is an improper access control issue in UniFi OS that could be exploited to make changes to vulnerable devices.

Global professional services company Accenture has acknowledged that it suffered a cybersecurity incident following claims by a threat actor that they stole "just over 35GB of source codes" from the company earlier this month. Several news outlets have reached out to Accenture, which has confirmed that the company is "aware of this isolated matter" and that the source of the breach has been remediated, but has not offered any additional information beyond saying that the incident did not affect Accenture operations and service delivery. In 2024, the same threat actor who posted about the incident attempted to sell Accenture employee data pilfered in a third-party breach. At the time, Accenture said that the data included just three Accenture-related names and email addresses. Accenture experienced another data security incident in 2021 following an attack by the LockBit ransomware group.