Hello, stay ahead with CISO Brief 🚀

🛡️ Researchers from ManageEngine, Google TAG, and Mandiant uncovered a supply-chain campaign that briefly modified AOMEI Backupper’s configuration files between April and July 2026 to serve a malicious ZIP and execute payloads silently. The attackers distributed a downloader named CROSSROAD and a backdoor called SHINYROAD via three short exposure windows, using the product’s promotion and notification mechanisms to bypass verification. Affected organizations were identified across multiple countries and sectors, and AOMEI says its CTO is investigating and has applied mitigations.

🔍 Cloudflare describes a focused effort to cut costs and speed up its AI code reviewer, lowering average review cost from around $2.46 to $0.77 and reducing median review time from 4m 09s to 1m 32s. Changes included code and architecture improvements, model selection adjustments, and cache/prompt engineering to increase prefix cache hits. They migrated to OpenCode v2, shared cached prefixes between reviewers, removed unnecessary tools, parallelized file reads, and tuned model effort levels. Model changes and effort tuning provided a larger share of savings than code changes, and targeted experiments validated that cheaper model variants maintained review quality for most workloads.

🚀 Amazon announces that CloudWatch Omni is now available in eight additional AWS Regions, extending its AI-powered observability experience beyond the initial regions. Omni unifies OpenTelemetry interoperability with CloudWatch scale, offering a standalone web experience with SSO and IDE extensions. It enables cross-account, cross-Region, and multi-cloud telemetry visibility, service discovery, dependency mapping, and golden metrics with chat and Agent Toolkit integration. Developers can install free Omni extensions for VS Code, Cursor, and Kiro to begin.

🧭 Cloudflare previews Clairic, a new System One extraction model hosted on Workers AI as @cf/cloudflare/clairic. Clairic extracts, computes, estimates, and forecasts typed values (numbers, dates, spans, lists, distributions) from unstructured text and images in a single forward pass, offering faster latency, lower cost, and higher accuracy than LLMs for structured-output tasks. It returns typed values or null with calibrated confidence and p_null, supports computed fields and intervals, and integrates with decision models and LLMs in pipelines.

⚙️ EC2 Image Builder can now use existing instances from an Auto Scaling Group (ASG) to perform image builds, reducing time spent launching and bootstrapping new instances. By specifying an ASG in your Image Builder infrastructure configuration, Image Builder selects an already running instance, applies build components, and creates the image while the ASG replenishes the instance pool. This opt-in feature doesn't change behavior for pipelines without an ASG and is available in all AWS Regions at no extra cost.

🛡️ Amazon GuardDuty now supports optional Runtime Monitoring detection rules covering process, file, and network activity. You can enable over 50 rules across EC2, EKS, and ECS (including Fargate) that map to 13 finding types and MITRE ATT&CK® tactics such as persistence and privilege escalation. Four new finding types are introduced and rules can be enabled in dry-run mode to validate detections before going live.

🧭 This article explains how MCP Apps (SEP-1865) extend the Model Context Protocol to pair backend tools with live UI workspaces inside Gemini Enterprise. It contrasts three assistant generations and describes responsive, visual workflows that run locally in a sandboxed iframe while staying synchronized with the assistant. The piece outlines security (OAuth, Cloud Run IAM), deployment steps, and how to register and enable a custom MCP Server as a Gemini Enterprise data store.

🔍 This post argues that LLMs used as evaluators are not inherently objective and can exhibit biases similar to humans. It outlines a three-pillar method to improve evaluation quality: internal stability checks via multi-pass variance testing, human alignment validation using chance-corrected metrics, and multi-judge ensembles with decomposed prompts and model ensembling. The piece highlights practical steps to detect inconsistency and drift in automated judges.

🛠️ This announcement introduces Python UDF logging in BigQuery, allowing developers to use Python's standard logging library inside UDFs to emit structured logs directly to Cloud Logging. The feature supports row-level diagnostic telemetry for fault-tolerant pipelines: malformed or corrupted rows can be quarantined (returned as NULL) while valid records proceed. Teams can create log-based metrics to trigger real-time alerts and query quarantined payloads via a linked _AllLogs view in BigQuery for triage and replay. The post walks through an e-commerce ingestion example and outlines best practices such as enabling debug output, avoiding sensitive data in logs, and sampling to manage costs.

🔍 A Moody's report found known exploited vulnerabilities take a median of 87 days to patch, leaving half unremediated after three months. The gap widens to 107 days for vulnerabilities without known exploitation evidence. The report warns that AI-driven discovery and exploitation are accelerating, increasing pressure on security teams and raising credit risk for organizations with weak patching practices.

🎯 Join Fortinet experts, IDC, and customers to examine how AI is reshaping threats and SecOps priorities at the Fortinet Security Operations Summit on November 4–5, 2026. The virtual two-hour program covers AI-powered threats, agentic AI for investigation and response, and Fortinet’s technology strategy, with demonstrations and practical guidance for CISOs and SOC teams. Sessions include threat intelligence, research insights, and a customer fireside chat.

🎮 Microsoft has resolved a known issue that caused games and Windows 11 apps using AC-3 (Dolby Digital) audio decoding to crash after installing the optional September 2026 preview update (KB5124010). The problem could cause applications to fail to launch or close unexpectedly, and affected apps included games, media players, and some productivity software. Microsoft says many modern apps that use their own decoders were not affected. The company recommends installing the October 2026 Patch Tuesday update (KB5127648) to address the issue on Windows 11 24H2 and later.

🔒 Conduent has agreed to an $80m class action settlement after a 2024 cyber incident that exposed personal and health data of roughly 62 million people. The funds will cover documented material losses, credit and medical monitoring for affected claimants, and plaintiffs' legal fees. The agreement splits claimants into two subclasses based on how accessible their stolen files were and caps direct cash reimbursement for documented fraud-related losses at $3,500 per claimant. Conduent denies liability, says it carries cyber insurance and will enhance data security as part of the settlement.

⚡ This week’s security roundup highlights high-impact incidents where trusted services, hardware supply chains, legacy systems, and AI agents were exploited. Major events include ConnectWise suspending cloud backup after unauthorized access, a BeyondTrust zero-day exploited by Armory Castle, Ledger reporting implanted backdoors in wallets, and Pwn2Own Ireland yielding 98 zero-days. The recap emphasizes the varied nature of attacks — from sophisticated espionage to trivial misconfigurations — and urges patching and stronger cyber hygiene.

🔎 Flare researchers detail how underground marketplaces sell complete Know Your Client (KYC) packages including forged documents, manipulated photos and videos, and human-operated verification to bypass identity checks. The report highlights services that prepare consistent identity data, fabricate supporting records, and use deepfakes or live operators for facial checks. Sellers also offer post-delivery account recovery and continued verification assistance to maintain control over compromised accounts.

🔒 F5 released out-of-band security patches to address multiple NGINX web server vulnerabilities, including three critical flaws that can enable remote code execution or denial-of-service. The issues affect the ngx_http_v3_module and the NGINX map directive and impact products such as NGINX Plus, NGINX Open Source, Ingress Controller, Gateway Fabric, and Instance Manager. F5 provided mitigation guidance for teams unable to immediately apply fixes, and noted no known in-the-wild exploitation so far.

🕵️ Federal investigators say executives at two ransomware negotiation firms maintained unusually close ties to the ShinyHunters extortion group, and one U.S.-based executive may have helped operate the group's victim pressure site. The probe follows the September breach of the FBI recruitment site and arrests linked to the group's leadership. Authorities are examining Signal messages seized during a Dutch arrest that appear to implicate multiple negotiators in sharing victim details and facilitating extortion.

🔒 OpenCode confirmed an October security breach affecting its Console service, potentially exposing API keys for OpenAI, Anthropic and Google AI models stored in some workspaces. The company reported the incident involved unauthorized access to a legacy system and impacted 1.4% of Console users, notifying affected customers by email. OpenCode said the breach affected the Console component only and has informed the relevant providers while advising remediation steps.