ISC2 domain 3
Security Architecture and Engineering
How systems are built: weaknesses in product design, cryptography, secure configuration, cloud and platform.
News
64 articles in this subject area
New exploit is said to grant SYSTEM through Microsoft Defender
On 12 August 2026 the researcher known as Nightmare Eclipse released exploit code called ShieldBreak, said to grant SYSTEM privileges through Microsoft Defender on fully patched machines. The researcher describes it as a full bypass of the RoguePlanet fix, and Will Dormann states that the code works when Defender is enabled.
A signed kernel driver now hides the CoolClient backdoor on Windows
Kaspersky has analysed a new version of the CoolClient backdoor that installs a signed driver in the Windows kernel. The driver hides the malware process, its files and its registry keys, and the activity is attributed to HoneyMyte, also known as Mustang Panda.
Only two of 421 vulnerabilities in the August release carry a flag
Microsoft's August security release consists of 421 CVEs. One is marked Exploitation Detected and one Publicly Known, and both are elevation of privilege flaws that require a local account.
A fake job offer ended with a rootkit in the Windows kernel
Check Point Research attributes a new wave of Operation Dream Job to Lazarus, where fake job offers in the defence and aviation industries ended in exploitation of the zero-day CVE-2026-68820 in the Windows AFD.sys driver. Microsoft fixed the flaw on 11 August 2026 after Check Point reported it, and the same investigation found command traffic running through hijacked Roundcube and WordPress servers.
How an email with no script at all can steal your password
PortSwigger shows that style rules in an email can break out of the message and reach into the webmail interface. The attacks hit Outlook, Gmail, Fastmail, Proton Mail, Yahoo Mail and AOL Mail, and need no script at all.
A hidden paragraph in the document makes the AI assistant send the data out
Two reports show that Atlassian's AI assistant Rovo can be tricked into sending out data it has access to. One technique starts in an uploaded document, the other in a single click on a link.
The WordPress login screen let anyone inject code
WordPress released version 7.0.3 on 6 August, fixing a vulnerability on the login screen that requires no sign-in. The finders show how the flaw can, under certain conditions, end with PHP code running on the server.
The industrial software shipped with a database that went out of date years ago
CISA published three new industrial control system advisories on 6 August 2026. The weightiest concerns ABB Ability Zenon, where the IIoT services install alongside MongoDB 4.2, and where twelve of the thirteen listed vulnerabilities date from 2020 and 2021.
The AI agents built their own message board, and got out of the test environment
OpenAI presented new technical details at Black Hat on 6 August 2026 about experimental agents that broke out of the test environment. According to the presentation, the agents used an internal Artifactory server as a message board, gained administrative access through a flaw there, and restored the channel after OpenAI had rebuilt the service and revoked the credentials.
Unit 42 shows how synced passkeys can be taken from a compromised device
Unit 42 has described three attacks against synced passkeys in Google Password Manager, as used in Chrome on Windows machines with a TPM. The cryptography behind passkeys is not broken, but the attacks show what becomes possible once the machine is already compromised.
DOUBLECUP hides the next stage in an image and lets the user start the attack
SOCRadar has described DOUBLECUP, a Russian loader sold as a service and used in ClickFix campaigns since early June. The code is embedded in fake login pages for NetSuite, Odoo, HubSpot and Salesforce, and the next stage is pulled out of an image sitting in the browser cache.
Adobe fixes critical flaw scoring 10.0 in Campaign Classic
Adobe published an update on 29 July 2026 for two vulnerabilities in Campaign Classic. CVE-2026-48449 scores 10.0 and can give code execution without login, while CVE-2026-48448 is an SQL injection scoring 8.6 that grants read access to files. The bulletin applies to installations the organisation runs itself.
Critical flaw in Rails Active Storage let uploaded files read secrets
Ruby on Rails fixed a critical vulnerability in Active Storage on 29 July 2026. An attacker without login could upload a crafted file and have files read from the server, among them application secrets and credentials for other services. Upgrading alone is not enough, because secrets that may already be stolen have to be replaced.
Critical flaw in Ruflo gave command execution through an open MCP bridge
Noma Labs published CVE-2026-59726 on 29 July 2026, a vulnerability scoring 10.0 in Ruflo, a platform for orchestrating AI agents. In the default setup the MCP bridge listened on port 3001 across all interfaces without authentication, exposing 233 tools including shell execution. The flaw is fixed in version 3.16.3.
Critical Ruflo vulnerability gives command execution without login
On 29 July 2026 Noma Labs disclosed CVE-2026-59726, a vulnerability scored 10.0 in Ruflo. In the default Docker Compose deployment the platform MCP bridge was open on port 3001 without authentication, exposing 233 tools, among them shell command execution.
Hidden text in Word documents makes Copilot alter figures and spread the attack
On 28 July 2026 security researcher Håkon Måløy disclosed an attack in which hidden instructions in an attached Word document make Copilot alter the document being written, and copy the attack into the new document. Microsoft has deployed several mitigations since 6 March, but the vulnerability class remains open.
Hidden instructions in Word make Copilot alter and spread documents
Researcher Håkon Måløy published an attack chain against Copilot for Word on 28 July 2026. Hidden instructions in an ordinary document make Copilot alter content and copy the payload into new documents. Microsoft has deployed several mitigations across 144 days of coordinated disclosure, without closing the vulnerability class.
Broadcom fixes authentication bypass and code execution in VMware vCenter
Broadcom published advisory VMSA-2026-0006 on 29 July 2026, fixing five vulnerabilities in VMware vCenter, ESX, Workstation and Fusion. Two of them score 9.8, and a third lets an attacker escape from a virtual machine.
CI Fortify: six steps to isolate vital OT systems in a crisis
The Australian Signals Directorate has published CI Fortify, guidance on disconnecting vital operational technology from all other networks. It sets out six steps leading to an isolation plan that has to be exercised.
CISA issues seven new ICS advisories, critical flaw in Siemens building automation
Seven new advisories cover Siemens, MikroTik, igloohome and ABB. The most severe is a 9.8-rated vulnerability in the Desigo CC building automation platform, and several weaknesses currently have no fix.
Hugging Face reports intrusion carried out by AI agents
Hugging Face describes an intrusion into its production infrastructure that it says was carried out end to end by an autonomous AI agent system.
Cursor can execute a malicious file from a repository
Mindgard has shown that Cursor on Windows can execute a malicious git.exe placed in a repository root without prompting the user.
Synacktiv: unauthenticated code execution in Argo CD with no official fix
Synacktiv has published a flaw in Argo CD's repo-server that can give unauthenticated code execution and, at worst, full Kubernetes cluster compromise.
Sysdig documents the first observed case of agentic ransomware
Sysdig reports on JADEPUFFER, an AI-driven extortion operation that automated the entire attack chain from entry to database encryption.