Hello, stay ahead with CISO Brief 🚀

🤖 This article from Google Cloud Consulting outlines five practical lessons for enterprises adopting agentic AI systems. It emphasizes redesigning workflows from relay-like handoffs to parallelized, agent-driven processes and shifting humans from step-level validators to outcome owners. The piece highlights trust-building through delegation practices, clear responsibilities, and low-risk experiments, while recommending hub-and-spoke governance and cross-team collaboration to accelerate safe, scalable adoption.

🛠️ Today Google AI & Infrastructure and Mercor introduce the Physical Engineers' Last Exam (PLE), a new benchmark and open competition to test agents on mechanical, electrical, civil, and structural engineering tasks. PLE provides 29 public tasks, grading tools, and a leaderboard at ple-bench.com, and includes long-horizon, real-world project folders with diverse file types. The benchmark evaluates code compliance, safety, and cost efficiency, revealing a wide gap between frontier models and professional engineering requirements.

🛡️ Google warns that threat actors are increasingly hijacking enterprise access to Large Language Models (LLMs) to fraudulently consume costly compute and AI services. Attackers obtain API keys via phishing, leaked code, misconfigurations, or exposed tokens to run expensive workloads while victims incur the charges. Incidents span multiple regions and industries, with single-day fraudulent charges exceeding $100,000 and cumulative losses reaching millions in some cases.

🛡️ BTCPay Server reported that attackers stole cryptocurrency from hot wallets on instances running version 2.4.1 or earlier, with about $1.63m traced to linked Bitcoin and Litecoin addresses. Blockchain analysis by Bitquery identified repeated sweeps between October 30 and November 2, and some activity continued after operators were warned to shut down. The project urged affected operators to rebuild servers with new wallets and seeds and noted that hardware wallets and wallets with spending keys held off-server were not impacted.

📰 Zero Wire is a new public vulnerability intelligence feed on Cloudflare Radar designed to give security teams concise, actionable briefings on critical, internet-facing vulnerabilities. Each card summarizes the issue, affected deployments, recommended actions, and links to vendor advisories and evidence. Briefings include contextual network observations when available, a suggested agent prompt for impact assessment, and an update chronology to track developments. The service complements vendor advisories and vulnerability databases rather than cataloging every CVE.

🛡️ Google outlines expanded measures to counter rising scam losses across the Asia Pacific region, which reached up to $114.1 billion USD in 2025. The company is deploying built-in protections such as Fake Call Detection and Pixel scam alerts, expanding Android developer verification, and promoting RCS-based defenses. Google is also engaging in industry coordination through the Industry Accord and new alliances like JASA, while Google.org commits $6 million to local partners for community training, AI prototyping, and policy support.

🔐 Late October’s emergency PHP patches for CVE-2026-109618 and CVE-2026-109619 aimed to stop remote code execution that could affect roughly 70% of websites. Despite fixes for multiple PHP branches, Censys found millions of vulnerable servers still unpatched two weeks later, while exploitation attempts and new exploits, including a related Brotli issue, continued to appear. Organizations must update PHP, audit systems, and watch for intrusion signs.

🧊 Supermarkets in at least nine European countries reported a cyber-attack that manipulated refrigeration systems, spoiling chilled and frozen stock. Retailers in Belgium, Estonia, Finland, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Sweden were affected, with temperatures reportedly set as high as 50°C. Authorities including Sweden’s NCSC and EU officials are investigating, while some UK food-supply businesses have also been impacted. The incidents mirror an October attack on refrigeration providers and raise concerns about deliberate disruption of food supply chains and potential nation-state involvement.

🛡️ Check Point Research (CPR) uncovered ‘prompt adware’ in a Codex jailbreak bundle called Haiou that persistently modifies an AI agent’s trusted natural-language instructions to inject advertisements for a third-party AI gateway. The finding shows how prompts, Skills and configurations — not executable code — can quietly alter agent behavior, creating visibility and supply-chain challenges for organizations. CPR warns this persistence mechanism could be used for more harmful manipulations, including steering recommendations or exfiltrating sensitive data, and advises firms to monitor agent instructions, vet third-party components and strengthen governance.

🔒 Microsoft telemetry shows the median vulnerability on enterprise network appliances remains unpatched for over a year, even as the most dangerous flaws are weaponized within days of disclosure. The analysis of roughly 383,000 CVEs through September 2026 found patch times vary widely: about two weeks for Windows workstations, more than a month for Windows Server, over 10 weeks for Linux and macOS applications, and more than a year for network appliances. Researchers warn that rapid exploit development — accelerated by AI — has shortened the defender’s response time, urging continuous monitoring, emergency fixes, and prioritization by exposure rather than CVSS alone.

🔒 The NCSC warns AI-powered malware is highly likely to be used in ransomware attacks on the UK’s critical national infrastructure by 2028, potentially affecting power stations, airports and hospitals. CEO Richard Horne urged organisations to address long-tolerated vulnerabilities and underinvestment in security, while the government announced major funding for national cyber programmes, AI security initiatives and SME support.

🔒 Red Hat has released fixed kernels for RHEL 8, 9, and 10 addressing a remote, unauthenticated IPv4 TCP use-after-free (CVE-2026-98432) that can lead to remote code execution; a reboot is required after update. Canonical has provided patches for Ubuntu 22.04, 24.04, and 26.04 in the -proposed testing pocket but has not set a regular release date. Red Hat and others published mitigations and guidance while upstream and other vendors already shipped fixes in October.

🛡️ ESET attributes a Russia-aligned actor called Gorbag to targeted campaigns against Ukrainian organizations since at least February 2025, favoring law enforcement, justice, local administrations near front lines, and defense. The group deploys lightweight information stealers such as GIFTEDCROOK and BeautyStealer to exfiltrate browser data, VPN configs, credentials, and documents, often via spear-phishing lures and malicious attachments. Observed tooling also includes Prime ReverseShell and DLL sideloading chains exploiting a patched WinRAR flaw (CVE-2025-8088), with stolen archives sent to Telegram or custom C2 servers.

🛡️ CrowdStrike and Microsoft reported that China-linked groups used an AI-powered penetration framework called ARTEX to probe and exploit security gaps across hundreds of organizations in October. The campaigns targeted governments and businesses, using multiple AI agents to automate reconnaissance and post-exploitation checks—sometimes identifying missing endpoint sensors and misconfigured cloud roles. Researchers warn that AI lowers the cost and time of attacks and urge improved visibility, credential hygiene, and continuous monitoring.

🛡️Google's Mandiant warns enterprises of a growing threat dubbed LLMJacking, where attackers bypass front-end chatbots to exploit backend large language model resources using exposed API keys and cloud tokens. Incidents this year span financial services, retail, technology and transportation, with fraudulent cloud charges exceeding $100,000 in a single day and up to $1m cumulatively before access was revoked. The ecosystem is maturing, with actors escalating privileges, abusing routing software, and establishing resale marketplaces that proxy and resell illicit LLM access.

🔍 AI, identity-based attacks, and expanding digital footprints are stretching cybersecurity teams thin, according to Barracuda’s Cybersecurity Skills Gap report. The survey of 2,000 IT and security decision-makers found that 67% say teams are struggling to scale to meet business needs while facing AI-related risks (68%) and advanced identity threats (67%). Organizations report reduced head count tied to AI and growing expectations to support digital operations without extra resources, prompting calls to combine AI with skilled human oversight.

🛠️ This post explains a five-rule engineering playbook Google Cloud uses to evaluate agent skills and avoid bloating context windows. It warns that single-shot testing and workstation benchmarks mislead evaluators, and describes tools like skill-reach, NVIDIA's SkillEvaluator, and sandboxed CI harnesses used to run paired ablations. Practical guidance covers routing, guardrails against reward-hacking, and what kinds of skills actually improve efficiency.

🔒 Synology is urging customers to update DiskStation Manager (DSM) after observing threat actors exploit a December 2024-patched vulnerability to deploy ransomware against devices running outdated DSM versions. The company referenced a November 2024 advisory listing three vulnerabilities (including CVE-2024-10441) and provided specific patched DSM version thresholds. Customers have reported infections attributed to C3RB3R ransomware that appends a ".L0CK3D" extension, and Synology recommends enabling automatic updates to reduce risk.