Researchers discovered a new social engineering technique that uses custom fonts and CSS to display malicious commands to a user while hiding them from AI assistants that parse the underlying HTML.
This vulnerability, which does not have a CVE identifier, affects numerous AI models by creating a discrepancy between what the user sees and what the AI analyzes, potentially leading to users executing harmful commands they believe have been verified as safe.
Vulnerability Details
Affected Product: Multiple AI Assistants including ChatGPT, Claude, Copilot, Gemini, Grok, and others.
Exploitation Status: A Proof-of-Concept (PoC) has been developed by researchers.
Risk & Impact
Triage: High; this technique bypasses security checks that users are beginning to trust, creating a significant risk of successful social engineering.
Attack Vector: An attacker creates a webpage with benign text in the HTML and a malicious payload that is encoded. A custom font and CSS are used to make the benign text invisible to the user and render the encoded payload as human-readable, malicious instructions (e.g., a command to create a reverse shell). The AI assistant, parsing only the HTML, sees the harmless text and incorrectly validates the page as safe.
Ease of Exploit: The technique does not require JavaScript or any browser vulnerabilities, relying only on custom fonts and CSS.
Action Plan
Immediate Action: Users should not solely rely on an AI's interpretation of a webpage to determine its safety. Always copy and paste the exact commands you are asked to run directly into the AI assistant for analysis.
Workaround: Be extremely cautious with any website that requests you to run commands, especially in a terminal or command prompt.
Detection: Security tools and AI vendors should work towards implementing render-and-diff analysis to detect discrepancies between the rendered page and the underlying Document Object Model (DOM).
Relevant professional terms
Font Rendering
The process by which a browser interprets font files and CSS to convert text within the HTML into a visual representation (pixels) for the user to see on screen.
Social Engineering
A manipulation technique used to trick users into making security mistakes or giving away sensitive information. In this context, it involves an AI assistant being used to falsely legitimize a malicious webpage.
A critical remote code execution (RCE) vulnerability in Microsoft SharePoint, identified as CVE-2026-20963, is now being actively exploited in the wild.
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has added the flaw to its Known Exploited Vulnerabilities (KEV) catalog, mandating federal agencies to apply patches.
Vulnerability Details
Affected Product: Microsoft SharePoint Enterprise Server 2016, SharePoint Server 2019, and SharePoint Server Subscription Edition.
Identifier: CVE-2026-20963
CVSS Score: 9.8 (Critical)
Exploitation Status: Actively Exploited
Risk & Impact
Triage: Urgent. CISA has ordered federal agencies to secure servers by March 21, 2026.
Attack Vector: An unauthenticated attacker can remotely execute arbitrary code on a vulnerable server through a network-based attack. The flaw is due to the deserialization of untrusted data and requires no user interaction to exploit.
Ease of Exploit: The attack complexity is considered low.
Action Plan
Immediate Action: Apply the security updates released during Microsoft's January 2026 Patch Tuesday.
Workaround: Organizations with end-of-support versions like SharePoint Server 2007, 2010, and 2013 should upgrade to a supported version to block attacks.
Detection: Monitor for signs of unauthorized code injection or remote execution on SharePoint servers.
Relevant professional terms
Remote Code Execution (RCE)
An attack in which a threat actor can remotely execute commands of their choosing on a target machine, effectively giving them control over that system across a network.
Deserialization of Untrusted Data
A vulnerability where an application deserializes data from an untrusted source without proper validation, which can lead to arbitrary code execution when malicious objects are processed.
The Interlock ransomware group is actively exploiting a zero-day vulnerability in Cisco Secure Firewall Management Center (FMC) Software.
This campaign targets enterprise networks, allowing attackers to execute arbitrary code with root-level privileges to deploy ransomware.
Key TTPs
Initial Access: Exploitation of a critical vulnerability (CVE-2026-20131) via crafted HTTP requests.
Execution: Arbitrary Java code execution as root, followed by fetching additional malicious binaries.
Defense Evasion: Use of legitimate remote access tools like ConnectWise ScreenConnect to blend in with normal traffic and maintain redundant access.
Campaign Analysis
This campaign highlights a significant threat, as Interlock leveraged a zero-day vulnerability for over a month before it was publicly disclosed, giving them a considerable advantage. An operational security failure by the threat actor exposed their entire toolkit, providing rare insight into their methods.
Targeting & Infrastructure
Target Profile: Critical infrastructure and businesses across North America and Europe using Cisco Secure Firewall Management Center software.
Relevant Terms
Zero-Day: A vulnerability in a system or device that has been disclosed but is not yet patched, which attackers can exploit.
Remote Code Execution (RCE): An attacker's ability to execute any commands of their choice on a remote target machine across a network.
Russian state-sponsored actor APT28 (aka Fancy Bear) is exploiting a cross-site scripting (XSS) vulnerability in Zimbra Collaboration webmail servers. The campaign targets Ukrainian government entities to steal credentials, exfiltrate emails, and gain long-term access to sensitive communications.
Key TTPs
Initial Access: Exploiting a reflected cross-site scripting (XSS) vulnerability (CVE-2022-27926) via crafted links in phishing emails.
Execution: Malicious JavaScript executes in the victim's browser to steal session cookies and credentials.
Defense Evasion: The attack payload is embedded directly within the HTML body of a phishing email, avoiding attachment-based detection.
Campaign Analysis
This campaign is part of APT28's broader, long-term strategy of targeting webmail platforms to conduct cyber-espionage. The group consistently leverages both new and older, unpatched vulnerabilities to maintain persistent access to high-value government and military targets.
Targeting & Infrastructure
Target Profile: Ukrainian government agencies and other European government, military, and diplomatic organizations.
APT (Advanced Persistent Threat): A stealthy and continuous computer hacking process, often orchestrated by a state-sponsored group, to steal data from a specific target.
XSS (Cross-Site Scripting): A type of security vulnerability where an attacker injects malicious scripts into trusted websites, which then execute in a victim's browser.
State-sponsored actors and commercial surveillance vendors are using the DarkSword exploit chain to compromise unpatched iPhones. The attacks, observed since late 2025, aim to steal a wide range of sensitive data, including cryptocurrency wallets.
Key TTPs
Initial Access: Victims visit compromised websites (watering hole attack) or malicious sites that contain a malicious iframe, requiring minimal to no user interaction.
Execution: A JavaScript-based exploit chain leverages multiple vulnerabilities, starting in the Safari browser to achieve remote code execution.
Defense Evasion: The malware performs a "hit-and-run" data exfiltration within minutes and then removes itself to evade detection.
Campaign Analysis
This campaign highlights the proliferation of sophisticated mobile exploit kits to a wider range of threat actors beyond elite state-sponsored groups. The rapid data theft and cleanup indicate a focus on efficient, hard-to-trace intelligence gathering and financial theft.
Targeting & Infrastructure
Target Profile: Users in Ukraine, Saudi Arabia, Turkey, and Malaysia have been targeted by various campaigns.
Infrastructure: Attackers use compromised websites, including government and news sites, as well as custom lure pages to deliver the exploit.
Relevant Terms
Exploit Chain: A sequence of vulnerabilities used by an attacker to gain progressively deeper access into a system, ultimately leading to a full compromise.
Watering Hole Attack: A cyberattack strategy where an attacker compromises a website that is frequently visited by a specific target group, rather than attacking the targets directly.
The U.S. and Israel launched "Operation Epic Fury," a coordinated military and cyber campaign targeting Iran's leadership, military, and government infrastructure. The operation integrates kinetic strikes with large-scale cyberattacks, aiming to disrupt and degrade Iranian strategic capabilities.
Key TTPs
Initial Access: SMS/phishing campaigns delivering mobile malware and exploitation of known vulnerabilities in internet-facing systems.
Execution: Deployment of wiper malware against financial and energy firms and hijacking of mobile applications for psychological operations.
Defense Evasion: Large-scale DDoS attacks and website defacements to create volumetric noise and distract defenders.
Campaign Analysis
This campaign signifies a major evolution in modern warfare, where cyber operations are fully integrated with kinetic military strikes from the onset. The initial phase caused a near-total internet blackout in Iran, severely degrading command-and-control while enabling psychological operations.
Targeting & Infrastructure
Target Profile: Iranian military leadership, IRGC command centers, government infrastructure, and nuclear facilities.
Infrastructure: Attacks targeted core internet infrastructure, including BGP routing and DNS, alongside SCADA/ICS systems.
Relevant Terms
Kinetic Warfare: Traditional military action involving physical force and conventional weapons, such as missile strikes and airstrikes.
Wiper Malware: A type of malicious software designed to permanently erase data from a targeted computer or network, often for destructive purposes rather than financial gain.
A global surge in mobile banking malware is now targeting over 1,200 financial applications, indicating a strategic shift by threat actors to commit fraud directly on user devices.
Key Findings
Research from Zimperium identified 34 active malware families targeting 1,243 financial apps across 90 countries.
The United States has the highest concentration of targeted apps, with 162 banking applications actively targeted.
Three malware families, TsarBot, CopyBara, and Hook, collectively target over 60% of the analyzed financial apps.
The Bottom Line
The proliferation of sophisticated mobile banking trojans signifies a critical shift in the threat landscape. Fraud has moved from centralized bank servers to the end-user's device, rendering traditional, perimeter-based security less effective.
Financial institutions must now prioritize on-device threat detection and robust in-app protection to counter attacks that originate from the compromised customer device itself.
Relevant Terms
Banking Trojan: A type of malicious software specifically designed to steal financial credentials from infected mobile devices, often by using fake login screens or intercepting 2FA codes.
On-Device Fraud: Illicit activities, such as unauthorized transactions or data theft, that are executed directly on a user's smartphone or tablet after it has been compromised by malware.
