Hello, stay ahead with CISO Brief 🚀

🔒 Model Context Protocol (MCP), the emerging plugin layer for agentic AI, has become a significant blind spot for security teams, introducing new shadow-AI risks much like shadow IT. CTEM programs can close this gap by extending scoping, discovery, prioritization, validation and mobilization to cover developer workstations, AI toolchains and MCP server configurations. Practical actions include actively enumerating MCP endpoints, scanning agent configuration and markdown context files for hardcoded API keys, and prioritizing exposures by attacker impact to produce actionable remediation tickets for engineering teams.

🔍 Sri Lankan police arrested 37 people, all Chinese nationals, on 2 May after raiding a property in Talangama, a Colombo suburb, following a tip-off. Officers seized 35 tablets, 147 mobile phones and 100 SIM cards and say several occupants were working illegally or overstaying visas. Authorities suspect romance-baiting operations that groom victims online and funnel them into fake cryptocurrency investment platforms. The arrests follow earlier large detentions and deportations tied to similar scam centres.

🔒 Penetration testing shows AI and LLM deployments contain a disproportionate share of severe vulnerabilities. Cobalt’s State of Pentesting Report finds 32% of LLM findings rated high risk versus 13% for legacy enterprise tests, and only 38% of those high-risk LLM issues are remediated. Experts point to emerging attack surfaces — notably prompt injection, now OWASP’s top LLM risk — broader blast radii from model integrations, and fragmented ownership for fixes. Recommended countermeasures include threat modeling, red teaming, least-privilege access, strict output validation, and human approval gates for high-consequence actions.

🔍 A healthcare customer bought servers in 2017 and, due to COVID-era lifecycle extensions and current supply-chain bottlenecks, now faces expiring vendor support and long lead times that prevent timely hardware refresh. The article recommends building a complete inventory using scanners (Nessus, Qualys, Rapid7, Greenbone/OpenVAS), network discovery (Nmap) and device fingerprinting (runZero), then mapping assets to NVD and CISA Known Exploited Vulnerabilities (KEV). Use a weighted risk formula to prioritize remediation and sort systems into immediate, managed, and monitored tiers. Document risk acceptance, deploy compensating controls where needed, and consider continuous monitoring with Wazuh.

🔒 Security researchers uncovered PCPJack, a credential‑theft framework that targets exposed cloud infrastructure and removes artifacts tied to TeamPCP. SentinelOne reports PCPJack worms through services to harvest credentials from Docker, Kubernetes, Redis, MongoDB, RayML and vulnerable web apps. Unlike many cloud campaigns it omits crypto‑mining and actively removes TeamPCP miner code, indicating monetization through credential theft, resale, fraud or extortion.

🔒 A jury found former federal contractor Sohaib Akhter guilty of conspiring to destroy dozens of government databases after being fired during a remote meeting in February 2025. Prosecutors say Akhter and his twin brother Muneeb ran write-protect commands and deleted roughly 96 databases hosting sensitive investigative and FOIA records for more than 45 agencies. They allegedly sought to hide their activity — even consulting an AI assistant about clearing system logs — and destroyed evidence; sentencing is set for September 9, 2026.

⚠ A newly disclosed Linux zero-day, named Dirty Frag, enables local attackers to obtain root privileges on most major distributions with a single command. Researcher Hyunwoo Kim published a detailed write-up and a proof-of-concept exploit after an embargo was broken on May 7, 2026. The flaw stems from an approximately nine-year-old logic error in the kernel's algif_aead interface and chains two page-cache write issues to modify protected files in memory. As a temporary mitigation, administrators are advised to disable and unload the esp4, esp6, and rxrpc modules until vendor patches are available.

🔒 A new unpatched local privilege escalation in the Linux kernel, called Dirty Frag, was disclosed to maintainers on April 30, 2026. Researcher Hyunwoo Kim (@v4bel) says it deterministically chains two page-cache write primitives (xfrm-ESP and RxRPC) to achieve root on many distributions, and a one-command PoC has been released. Vendors recommend immediately blocklisting the esp4, esp6, and rxrpc modules and monitoring upstream and vendor advisories for patches.

🔒 Instructure's Canvas platform was taken offline on May 7 after the cybercrime group ShinyHunters defaced login pages and posted a ransom demand claiming to hold data on 275 million students and faculty at nearly 9,000 institutions. Instructure had acknowledged a breach on May 6, saying the stolen records include names, email addresses, student ID numbers and user messages but not passwords or financial information. The outage, timed during many institutions' final exams, disrupted coursework while schools and the vendor evaluated exposure and potential extortion responses.

⚠️ Palo Alto Networks has confirmed a critical zero-day in PAN-OS's Captive Portal (CVE-2026-0300) that allows unauthenticated remote code execution as root on exposed PA and VM series firewalls. Reporting indicates suspected state-sponsored actors exploited the flaw for nearly a month. Palo Alto plans updates beginning May 13; customers should restrict or disable the portal until patches are available.

🔒Google has increased maximum payouts for its vulnerability reward programs, raising the top prize to $1.5 million. The new maximum applies to critical issues impacting Android, with reports indicating the full amount requires compromising the Pixel Titan M2 security chip. Rewards for vulnerabilities in Chrome now top out at $250,000. Since launching its programs in 2010, Google has paid $81.6 million to researchers.

⚠️ Thirteen critical vulnerabilities have been disclosed in the vm2 JavaScript sandbox, including a full sandbox escape (CVE-2026-26956) that can allow attacker-controlled code to execute host commands under specific Node.js 25/WebAssembly conditions. Another high-risk issue (CVE-2026-44007) involves NodeVM nesting interacting with the legacy module resolver and was patched in 3.11.1. Developers should upgrade to vm2 3.11.2 immediately and consider interim mitigations such as avoiding Node 25 runtimes or disabling WebAssembly for untrusted sandboxes.

🔒 The ShinyHunters extortion group defaced Canvas login portals for roughly 330 colleges and universities, replacing standard pages with an extortion message that demanded payment by May 12, 2026. The same message also appeared in the Canvas app and was visible for about 30 minutes before being taken offline. Instructure has taken Canvas offline while confirming that data was stolen and continuing its investigation. BleepingComputer reports the group claims the theft includes extensive student and staff records.

🔐 At the World Economic Forum Annual Meeting on Cybersecurity 2026, Fortinet highlighted how AI and deepfakes are reshaping attack surfaces, with identity now a primary vector and attackers operating in structured, continuous campaigns. Discussions stressed that AI accelerates reconnaissance and exploitation while defenders contend with fragmentation, governance gaps, and inconsistent visibility. Fortinet urged platform consolidation, stronger identity and exposure management, and operationalized public-private collaboration to better align detection with response.

⚠️ A new banking trojan named TCLBanker is being distributed via a trojanized MSI installer for Logitech AI Prompt Builder and targets 59 banking, fintech, and cryptocurrency platforms, with initial activity observed mainly in Brazil. Researchers at Elastic Security Labs report the malware uses DLL side-loading and strong anti-analysis defenses, runs persistent watchdogs to detect debuggers, and monitors the browser address bar to trigger theft routines. It provides remote-control capabilities (live streaming, screenshots, keylogging, clipboard theft, and shell execution) and uses WPF overlays to capture credentials. Uniquely, TCLBanker includes worm modules that hijack WhatsApp Web sessions and abuse Microsoft Outlook to self-propagate to contacts, increasing the risk of rapid spread.

🔒 Palo Alto Networks introduces Frontier AI Defense, a platform initiative designed to counter next-generation, agentic AI threats that can autonomously discover and chain software flaws. Their testing of frontier models (including GPT-5.5-Cyber, Mythos, and Claude Opus 4.7) revealed a step-change in coding capability and attack automation. The program combines Unit 42 expertise, early model access, platform integration, and partner alliances to enable prioritized mitigation and autonomous remediation at machine speed.

🦙 A critical vulnerability in Ollama (CVE-2026-7482) allows unauthenticated attackers to upload a crafted GGUF model file and trigger an out-of-bounds heap read in the model quantization pipeline. The flaw can leak process memory — including system prompts, conversation history, environment variables, API keys, and other secrets — to remote servers. Update to Ollama 0.17.1 and restrict network access.

🚀 Amazon Web Services now offers Amazon EC2 G6 instances powered by NVIDIA L4 GPUs in the AWS European Sovereign Cloud (Germany). These instances address graphics-intensive and machine learning workloads — including natural language processing, translation, video and image analysis, speech recognition, and personalization — with up to 8 L4 Tensor Core GPUs (24 GB each), third-generation AMD EPYC processors, up to 192 vCPUs, 100 Gbps networking, and 7.52 TB local NVMe SSD. G6 instances are available as On-Demand, Spot, and Savings Plans and can be launched via the AWS Management Console, CLI, or SDKs. They expand AWS's GPU capabilities for customers with sovereignty and compliance needs.