Hybrid Cloud
Cyber Security & NaaS
AI & Data
IoT & Connectivity
Business Applications
Intelligent Workplace
Consulting & Professional Services
Small Medium Enterprise
Health and Social Care
Industry
Retail
Tourism and Leisure
Transport & Logistics
Energy & Utilities
Banking and Finance
Smart Cities
Public Sector
Cybersecurity Weekly Briefing, 7-13 March
PROMPTFLUX uses Gemini's API to rewrite its own code in real time during an attack
Mandiant has identified a significant evolution in the use of artificial intelligence in cyberattacks, with state actors and criminals integrating LLM and AI APIs directly into malware and attack infrastructure.
Throughout 2025, AI went from being used for auxiliary tasks such as phishing generation or code assistance to becoming an operational component. Notable examples include PROMPTFLUX, which uses the Gemini API to dynamically rewrite its own code through real-time modifications, and PROMPTSTEAL, linked to APT28, which queries language models to generate Windows commands for document theft. Other implants such as FRUITSHELL and QUIETVAULT incorporate AI capabilities to locate sensitive data and automate exfiltration.
The report also warns of the growing risk of Shadow AI in companies, where AI tools are deployed without security controls, creating problems with visibility, asset management and access control.
Chrome 146 fixes 29 vulnerabilities, including the critical CVE-2026-3913 flaw in WebML
Google has released version 146 of Chrome, fixing 29 vulnerabilities, including CVE-2026-3913 (no CVSS yet but considered critical by Google), a heap buffer overflow in WebML.
The patch also addresses 11 high-severity flaws, including buffer overflows (CVE-2026-3915), integer overflows (CVE-2026-3914) and multiple use-after-free conditions in components such as Agents, WebMCP, Extensions, TextEncoding, MediaStream, WebMIDI and WindowDialog. Out-of-bounds access flaws in Web Speech and WebML are also fixed. The set is rounded out with 17 medium- and low-severity vulnerabilities affecting V8, Skia, PDF, DevTools, and other modules, including insufficient policy issues, out-of-bounds reads, and security interface errors.
It is recommended to immediately update to Chrome 146.0.7680.71 for Linux and 146.0.7680.71/72 for Windows and Mac to mitigate potential attacks.
BeardShell and Covenant drive a new wave of advanced espionage by APT28
ESET researchers have identified the reactivation of APT28, also known as Fancy Bear, with a modern arsenal based on two paired implants: BeardShell and Covenant, which use different cloud providers to ensure resilience and persistence in espionage operations, especially against Ukrainian military personnel.
Between 2025 and 2026, APT28 deployed BeardShell alongside a deeply modified Covenant for the long term, with new protocols based on cloud services such as Filen, Koofr and pCloud. The structural changes in Covenant, including a deterministic machine identification system and modifications to the execution flow to evade detection, demonstrate an active and sophisticated development capability.
This resurgence indicates that APT28's advanced implant infrastructure did not disappear, but evolved quietly, likely reactivated by the conflict in Ukraine and kept off the radar of analysts until now.
Iranian cyber doctrine combines internal digital isolation and covert operations through state-sponsored APTs
According to a report by FalconFeeds, Iran's cyberoperational landscape in 2026 is characterised by a combination of mature offensive operations and a defensive doctrine based on digital isolation through the National Information Network (NIN), which employs selective blackouts, BGP/DNS manipulation and protocol suppression (HTTP/3, IPv6) to reduce the attack surface and control dissent.
This strategy limits the international visibility of its activity, but not its capabilities, which are complemented by the use of alternative solutions such as commercial satellite links. In parallel, the Ministry of Intelligence and Security (MOIS) maintains ongoing technical espionage operations through MuddyWater, which abuses compromised accounts, legitimate RMMs and commercial VPNs, and Prince of Persia, which relies on malware such as Foudre v34 and Tonnerre v50. For its part, the Islamic Revolutionary Guard Corps (IRGC) operates offensive clusters such as OilRig/APT34, characterised by the abuse of cloud credentials and DNS tunnelling, and Charming Kitten/APT35, which develops its own frameworks and exploits known vulnerabilities.
In addition, related actors such as Handala function as narrative multipliers through hack-and-leak campaigns and exaggeration of impact.
Attack via Microsoft Teams to deploy A0Backdoor in financial and healthcare entities
Researchers at BlueVoyant have identified a campaign targeting organisations in the financial and healthcare sectors that uses social engineering via Microsoft Teams to gain remote access using Quick Assist.
The attacker initiates the operation by flooding the target's email with spam and then impersonates IT support to convince the victim to start a remote session. Once access is gained, digitally signed MSI installers are deployed that simulate Teams or CrossDeviceService components. The intrusion uses DLL sideloading with a malicious library (hostfxr.dll) that decrypts shellcode in memory and deploys A0Backdoor, encrypted with AES and protected by sandbox detection.
The malware collects system information using Windows APIs and establishes C2 communication through DNS MX queries with encoded subdomains, a technique designed to evade DNS tunnelling detection controls. The campaign shows strong similarities to BlackBasta's tactics.
◾ This newsletter is one of the deliverables of our Operational and Strategic Intelligence service. If you are interested in knowing the rest of the Operational and Strategic Intelligence contents included in the service, please contact us →
