The Cyber Byte - 10 March 2026

This edition highlights the proliferation of sophisticated, likely nation-state-developed iOS exploit kits into the hands of criminal actors for mass campaigns, alongside the disclosure of a new class of zero-click vulnerabilities in agentic AI browsers that enable silent data and credential theft. State-aligned actors are accelerating operations, with Iranian groups targeting IP cameras for battlefield intelligence and threat actors from Pakistan and China leveraging AI to rapidly develop and deploy offensive tooling. These trends indicate a rapidly evolving threat landscape where advanced capabilities are becoming more widespread, and AI is increasingly integrated into both offensive and defensive cyber operations.

Significant Cyber Incidents and Articles of Interest

• Mass iOS Exploitation via Leaked Nation-State Framework: A sophisticated iOS exploit kit, named "Coruna" by its developers and dubbed "CryptoWaters" by security firm iVerify, is being used in the first observed mass exploitation of iPhones [8, 9]. The framework, which iVerify assesses has similarities to tools developed by US government-affiliated actors, was deployed by a financially motivated Chinese threat actor (UNC6691) in widespread watering hole attacks targeting visitors of pornography and cryptocurrency websites. The exploit kit targets a wide range of iOS versions (13.0 to 17.2.1) and delivers a payload called PLASMAGRID that is designed to steal cryptocurrency and harvest sensitive data from numerous financial and messaging applications. [8, 9].

• "PleaseFix" Vulnerabilities Disclosed in Agentic AI Browsers: Zenity Labs disclosed "PleaseFix," a family of critical vulnerabilities affecting agentic AI browsers, including Perplexity Comet, that permit zero-click agent hijacking [7]. The vulnerability stems from the browser's agent inheriting user permissions to autonomously execute tasks. Attackers can embed malicious content in routine workflows, such as a calendar invite, which triggers the agent to access the local file system for data exfiltration or manipulate authenticated sessions with password managers like 1Password to steal credentials, all without user interaction or awareness. This discovery reveals a fundamental flaw in the emerging agentic computing model, where extending user trust to automated agents introduces new attack vectors that bypass traditional security controls [7].

• Mobile Spyware Campaign Impersonates Israeli "Red Alert" App: A targeted smishing campaign is distributing a trojanized version of the "Red Alert" rocket warning Android application to Israeli citizens [4]. The SMS messages, discovered on March 1, 2026, impersonate official Home Front Command communications and trick users into installing the malicious app. While the app retains its full, legitimate rocket-alert functionality to avoid suspicion, it covertly operates as spyware in the background, collecting sensitive data including SMS messages, contact lists, precise location data, and device account information. Acronis assesses with medium confidence that the campaign may be linked to the threat actor Arid Viper (APT-C-23), highlighting the weaponization of trusted emergency services during periods of geopolitical conflict to maximize social engineering effectiveness [4].

• Study Reveals Thousands of Valid Certificates Exposed by Leaked Private Keys: A joint research initiative between Google and GitGuardian has quantified the systemic risk posed by private keys leaked on public code repositories [5]. The study analyzed approximately one million unique private keys found on GitHub and DockerHub since 2021 and successfully mapped them to 140,000 real-world TLS certificates. As of September 2025, 2,622 of these certificates were still valid, exposing major organizations, including Fortune 500 companies and government agencies. The research revealed a widespread failure in remediation and revocation processes, with only a 9% response rate from affected organizations and many Certificate Authorities failing to revoke compromised certificates even after being contacted [5].

• OpenAI Codex Security Launch: OpenAI released Codex Security, an AI-powered application security agent designed to identify complex vulnerabilities with high confidence while reducing false positives. The platform has already demonstrated significant impact by discovering critical vulnerabilities in major open-source projects, resulting in 14 assigned CVEs, including critical flaws in OpenSSH, GnuTLS, GOGS, and PHP. Over the past 30 days, Codex Security scanned more than 1.2 million commits, identifying 792 critical findings and 10,561 high-severity findings while maintaining critical issues in under 0.1% of scanned commits. The platform's ability to build system-specific threat models and provide automated validation in sandboxed environments represents an advancement in defensive capabilities, though the underlying AI technology could potentially be adapted for offensive purposes [2].

Threat Actor Activity

• Iran-Nexus Actors: Iran-nexus threat actors have intensified targeting of IP cameras across Israel, Qatar, Bahrain, Kuwait, UAE, Cyprus, and Lebanon since February 28, 2026, using this reconnaissance for battle damage assessment and missile operation support. The campaign exploits vulnerabilities in Hikvision and Dahua cameras (CVE-2017-7921, CVE-2021-36260, CVE-2023-6895, CVE-2025-34067, CVE-2021-33044) through infrastructure combining commercial VPN exit nodes and virtual private servers attributed to multiple Iran-nexus actors. This activity demonstrates Iran's doctrine of leveraging camera compromise for operational support prior to kinetic strikes, with similar patterns observed during the 12-day Israel-Iran conflict in June 2025. The integration of cyber and physical warfare capabilities poses significant risks to critical infrastructure monitoring and early warning systems across the region [1].

• APT36 (Transparent Tribe): The Pakistan-based group APT36 has shifted to an AI-assisted development model, termed "vibeware," to industrialize the production of a high volume of mediocre malware implants [6]. This approach allows the actor to rapidly generate disposable tools in niche programming languages like Nim, Zig, and Crystal to evade signature-based detections. APT36 is also heavily leveraging a "Living Off Trusted Services" (LOTS) strategy for command and control, abusing legitimate platforms such as Slack, Discord, Supabase, and Google Sheets. The group's primary targets remain the Indian government and its foreign embassies, with initial access achieved through phishing emails containing malicious shortcut files [6].

• China-Nexus Actors: A China-based developer ("Ed1s0nZ") with assessed ties to the Chinese Ministry of State Security (MSS) has developed an open-source, AI-native offensive security tool named CyberStrikeAI [3]. The tool, which integrates over 100 security utilities, is seeing increased adoption, with 21 unique IP addresses observed hosting the platform between late January and late February 2026, primarily in China, Singapore, and Hong Kong. The developer's interactions with MSS-affiliated firms like Knownsec and the CNNVD suggest the tool may be leveraged by Chinese state-sponsored APTs for operations, including reconnaissance and targeting of vulnerable edge devices like Fortinet FortiGate appliances [3].

• UNC6691: This financially motivated threat actor, operating from China, has acquired and deployed the sophisticated Coruna iOS exploit kit in large-scale watering hole campaigns [9]. The group uses fake Chinese-language websites related to finance and cryptocurrency to lure victims. Once a user visits on a vulnerable iOS device, the exploit chain is delivered, ultimately deploying the PLASMAGRID payload. This implant is designed to steal financial information by hooking into numerous cryptocurrency wallet applications and exfiltrating sensitive data like backup phrases [9].

Indicators to Watch

• AI-Driven and Niche Language Malware: Malware generated with AI assistance, often written in less common languages like Nim, Zig, Rust, and Crystal, to bypass traditional defenses. APT36's "vibeware" suite includes tools such as Warcode, CrystalShell, SupaServ, and LuminousStealer [6]. Monitor for the execution of unsigned binaries, particularly those initiating suspicious network connections or system calls [6].

• Living Off Trusted Services (LOTS) for C2: Threat actors are increasingly abusing legitimate cloud services for command and control. Security teams should monitor for anomalous or persistent outbound traffic to platforms like Discord, Slack, Google Sheets, Supabase, and Firebase from non-standard applications, as this is a core TTP for groups like APT36 [6].

• Proliferation of Offensive AI and Exploit Kits: Offensive tools like CyberStrikeAI and exploit kits like Coruna are becoming increasingly accessible to threat actors of all skill levels. Organizations should monitor for suspicious server banners on open ports and implement defenses against watering hole attacks targeting high-risk websites across multiple platforms [3, 9].

• Targeting of Edge and IoT Devices: Iranian actors are actively exploiting vulnerabilities in internet-facing IP cameras (Hikvision and Dahua). This activity can serve as an early warning indicator for potential follow-on kinetic military action in the Middle East. Organizations should ensure such devices are patched, segmented, and not directly exposed to the internet [1].

Key IOCs:

• Android Spyware (Red Alert): Package name com.red.alertx and C2 domain ra-backup[.]com [4].

• CyberStrikeAI Servers: A list of 21 IPs has been published by Team Cymru, including 106.52.47.65, 115.120.233.95, and 47.95.33.207 [3].

• Coruna/PLASMAGRID C2 Domains: A large list of DGA-generated .xyz domains has been identified, including vvri8ocl4t3k8n6.xyz, rlau616jc7a7f7i.xyz, and ol67el6pxg03ad7.xyz [9].

• Targeted Vulnerabilities (Iran): CVE-2017-7921, CVE-2021-36260, CVE-2023-6895, CVE-2025-34067 (Hikvision), and CVE-2021-33044 (Dahua) [1].