SAGE INTELLIGENCE BRIEF Sunday, August 02, 2026 =========================================== LEAD STORY Storm-2945, an SVR sub-cluster, has been hijacking hotel Wi-Fi captive portal DNS since May to deliver CornFlake, a Go-based RAT that pulls Chrome App-Bound Encryption-protected credentials and harvests M365, Azure AD and WAM tokens without ever touching a browser cookie. Since July 16 the operation pivoted to Microsoft device-code authentication flow, meaning a traveling employee who enters an attacker-supplied code on Microsoft's legitimate sign-in page hands over MFA-satisfied access. For any enterprise with Entra tenants and road warriors, this is an active extraction operation targeting your identity plane right now. --- CONNECTING THE THREADS The Storm-2945 hotel Wi-Fi operation connects directly to the control-plane identity thread I've been tracking since the Sapphire Sleet npm compromise and the ShinyHunters/Salesforce guest account incidents. Every one of these attacks routes around endpoint controls and targets session state, tokens or trusted-service credentials. The pattern is consistent: perimeter and endpoint controls are being stepped over entirely, and identity is the actual battleground. The device-code flow pivot is the sharpest signal yet, because the victim completes authentication on Microsoft's own infrastructure and the attacker gets the session anyway. The Coldcard seed entropy failure and the Adform ad-script poisoning belong to the same structural category I flagged when the CosmosEscape Cosmos DB story landed: isolation guarantees that exist at a layer above the substrate can be defeated by attacking the substrate itself. Coldcard's entropy weakness was in the PRNG below the wallet's security model. Adform's malicious blocks rode inside a trusted first-party script. CosmosEscape recovered signing keys from multi-tenant gateway nodes. In every case the security perimeter held, but the foundation under it failed. The durable lesson that's been compounding all month: the layer below your trust boundary is always the attack surface. The Adobe Campaign Classic CVSS 10.0 zero-interaction RCE follows the pattern I flagged after the vCenter unauthenticated CVE last night. When there's no user interaction required and no workaround available, patch velocity is the only meaningful control variable. The vCenter story confirmed that network placement of management infrastructure doesn't substitute for patching. Tonight's ACC story confirms the same principle extends to marketing automation infrastructure, which typically sits inside the enterprise perimeter with access to customer data pipelines and CRM integrations. --- IT INFRASTRUCTURE ARCHITECTURE Windows ISO Bloat Driven by AI Components Microsoft's Windows installation files are growing in size as AI components get bundled into the base image. For MSPs managing deployment pipelines, storage provisioning for imaging infrastructure and slower-than-expected provisioning times on drives under 256GB are both operational realities now. Source: https://www.zdnet.com/article/windows-installation-files-getting-bigger-blame-ai/ Enterprise Cloud Spend Clears $143B Per Quarter Cloud infrastructure revenue is now above $143 billion per quarter with growth accelerating. For any client still treating cloud as a transitional state rather than the primary compute model, the economics of on-premises refresh as an alternative need to be defended, not assumed. Source: https://www.theregister.com/off-prem/2026/08/01/enterprise-cloud-infrastructure-uptake-shows-no-sign-of-slowing/5281835 AWS Builder Center Free Sandbox Environments AWS now offers free, time-limited sandbox environments through Builder Center for workshops and training. The practical value for MSPs is eliminating the friction of standing up isolated lab accounts for customer enablement sessions or internal training, with no billing exposure. Source: https://www.infoq.com/news/2026/08/aws-builder-sandbox/ --- CYBERSECURITY & COMPLIANCE Operation CaptiveCrunch: Hotel Wi-Fi as an Identity Extraction Platform Storm-2945 has been running DNS hijacks on hotel captive portals since May to deliver CornFlake RAT and ChocoShell, harvesting M365, Azure AD and WAM tokens via .tbres file reads and enabling session replay. The device-code authentication pivot since July 16 is the critical escalation. Blocking device-code flow in Conditional Access and enforcing always-on full-tunnel VPN for travelers are the two immediate controls. Source: https://thehackernews.com/2026/08/hijacked-hotel-wi-fi-pushes-fake.html Adobe Campaign Classic CVSS 10.0, CVE-2026-48449 Zero-interaction arbitrary code execution in Adobe Campaign Classic, patched only in v7 build 7.4.3 / build 9398. A companion SQL injection bug, CVE-2026-48448 at CVSS 8.6, enables arbitrary file reads. Confirm your exact build line before attempting to apply because earlier build lines have no patch path. Source: https://thehackernews.com/2026/08/adobe-campaign-classic-cvss-100-flaw.html Adform Ad Script Poisoned to Swap Crypto Wallet Addresses Attackers appended two XOR-obfuscated malicious blocks to trackpoint-async.js served from s2.adform.net, polling clipboards every four seconds and hooking DOM value setters on input elements to replace Bitcoin, Ethereum and Tron addresses in real time. Any enterprise or client running Adform tags needs to audit whether the compromised script version touched their properties and assess whether customer-facing transaction flows were exposed. Source: https://thehackernews.com/2026/08/hackers-poison-adform-script-to-swap.html Coldcard Hardware Wallet PRNG Flaw, $70M Drained in 41 Minutes A 2021 firmware error caused Coldcard to call a software PRNG instead of the STM32 hardware RNG, yielding 40-72 bits of effective entropy against the 128 bits BIP-39 requires. Emergency firmware shipped July 31, but the fix does nothing for existing wallets. Any seed generated on affected firmware must be regenerated on patched firmware and funds moved. Restoring the old seed carries the vulnerability forward regardless of which device it's restored to. Source: https://thehackernews.com/2026/08/coldcard-hardware-wallet-flaw-linked-to.html --- CLOUD PLATFORMS & STRATEGY CosmosEscape: Azure Cosmos DB Primary Key Exposure The CosmosEscape vulnerability exposed primary keys for Cosmos DB accounts from multi-tenant gateway nodes, granting full read-write access. This is a direct confirmation of the principle I flagged when this first landed: network-isolated cloud database designations aren't equivalent to physical network segregation. Workloads with hard data-boundary requirements need controls outside the managed service layer. Source: https://www.securityweek.com/ No additional unique developments tonight beyond what's covered in Cybersecurity and Infrastructure. --- NETDEVOPS & NETWORK AUTOMATION No notable developments tonight. --- AI IN INFRASTRUCTURE & AIOPS Kaseya Shifts to Open Agentic AI Framework for MSPs Kaseya previewed an open, API-driven agentic framework for Kaseya Intelligence at Connect Europe 2026, moving away from closed ecosystem tooling. For MSPs evaluating AI-assisted RMM and PSA workflows, the signal is that the architecture is shifting toward composable agent pipelines rather than monolithic AI features, which changes how integration work should be planned. Source: https://www.technewshub.co.uk/msp-news AI in F1: The Human-in-the-Loop Advantage Aston Martin Aramco's experience with AI in Formula One points to practitioner judgment as the variable that determines whether AI tooling produces competitive advantage. The same pattern applies to AIOps: teams that keep a skilled operator in the decision loop outperform teams that automate the judgment layer away. Source: https://www.zdnet.com/article/ai-in-formula-one-competitive-advantage-is-all-about-the-human-in-the-loop/ --- HARDWARE, GPU & COMPUTE Nvidia Vera CPU Deep Dive: 88 Olympus Cores, 1.5TB of Memory Nvidia's Vera CPU uses 88 custom Olympus cores with 176 threads, 1.5TB of on-package memory and 1.8TB/s of NVLink connectivity. It's purpose-built for AI inference and training workloads at a scale that makes it a data center fabric component, not a general-purpose processor. The NVLink bandwidth figure alone reframes what host-to-accelerator data movement looks like in next-generation AI infrastructure. Source: https://www.theregister.com/systems/2026/08/01/nvidias-vera-cpu-and-the-olympus-cores-that-power-it-deep-dive/5282056 Kioxia CM10 PCIe Gen6 SSDs Now Available Kioxia's CM10 series spans 2.5-inch and EDSFF form factors, covers both PCIe Gen5 and Gen6 interfaces and offers air-cooled and liquid-cooled configurations. Gen6 SSD availability matters for storage architects planning infrastructure that needs to keep pace with GPU memory bandwidth in AI training clusters. Source: https://www.servethehome.com/kioxia-cm10-series-launched-for-the-pcie-gen6-generation-of-ssds/ --- NETWORK MANAGEMENT & MONITORING CyberProof Agentic MXDR Platform Launch CyberProof launched an agentic MXDR platform targeting enterprise SOCs, with AI agents handling up to two-thirds of security investigations autonomously under human governance. For MSPs evaluating MSSP capability expansion, this establishes the competitive floor. SOC-as-a-service offerings without some agentic investigation layer are already behind the operational baseline larger players are shipping. Source: https://www.technewshub.co.uk/msp-news --- MANAGED SERVICE PROVIDERS MSP-to-MSSP Signal: Kaseya and CyberProof Both Moving on Agentic SOC Two separate platforms announced agentic AI SOC capabilities this week. The direction is clear: managed security is getting restructured around AI-assisted investigation workflows, and MSPs that haven't mapped their security service evolution to this architecture are going to face margin compression as agentic MXDR becomes the baseline expectation from enterprise buyers. Source: https://www.technewshub.co.uk/msp-news --- IT VENDOR ECOSYSTEM & M&A Windows Activation Error Surfaces in Airport Check-In Infrastructure A Windows Activation error appeared in check-in kiosk infrastructure at an airport, surfacing the persistent risk of unmanaged Microsoft licensing in embedded and kiosk deployments. For MSPs managing retail or hospitality clients with Windows-based POS or kiosk fleets, licensing state validation belongs in the monitoring stack. Source: https://www.theregister.com/offbeat/2026/08/01/unexpected-item-in-the-bagging-area-as-windows-activation-error-pops-up-at-check-in/5281946 --- EDGE COMPUTING & IOT AMOS MacOS Stealer Active in the Wild SANS ISC flagged an active Atomic MacOS stealer infection chain this weekend. For MSPs managing mixed-OS environments with macOS endpoints at client sites, particularly in creative, executive or engineering roles, endpoint detection coverage on macOS needs the same validation cadence applied to Windows fleets. Source: https://isc.sans.edu/diary/rss/33208 Phishing Campaigns Now Targeting AI Solutions Providers SANS ISC documented phishing campaigns specifically targeting AI solutions providers, exploiting fear of losing access to AI tooling or credentials. As clients accelerate AI adoption and staff use more AI-adjacent SaaS, the phishing surface expands proportionally and staff need explicit conditioning around AI platform credential solicitation. Source: https://isc.sans.edu/diary/rss/33206 --- SALES & REVENUE Know When to Walk Away Before the Conversation Starts In "The Art of the Deal" by Roger Dawson ("Secrets of Power Negotiating"), Dawson's core argument is that your walk-away position must be defined in advance, not calculated under pressure. Most salespeople set their floor during the negotiation itself, when anchoring, time pressure and relationship investment have already compromised their judgment. Setting the minimum acceptable outcome in writing before the first call is a structural protection against giving away margin you didn't intend to give. Source: "Secrets of Power Negotiating" by Roger Dawson (Goodreads compounding) The Follow-Up Cadence Is Where Most Revenue Gets Left Behind "New Sales. Simplified." by Mike Weinberg makes the point that most deals are lost to silence after the first conversation. Weinberg's data shows salespeople follow up an average of 1.3 times before abandoning a prospect, while most enterprise deals require 6-8 touches to advance. Building a structured, calendar-governed follow-up sequence into the sales process rather than relying on memory or CRM reminders is where the gap between average and top performers shows up most visibly. Source: "New Sales. Simplified." by Mike Weinberg (Goodreads compounding) --- REAL ESTATE & INVESTMENT Debt Service Coverage Is the Variable That Determines Survivability In "The Complete Guide to Real Estate Finance for Investment Properties" by Steve Berges, the DSCR calculation is framed as the primary indicator of whether a property can service its debt through a vacancy cycle without requiring capital infusion from the owner. Underwriting to a 1.0 DSCR is a breakeven posture, not a safety margin. Disciplined investors target 1.25 or higher to absorb a 15-20% revenue interruption without a distress event. Source: "What Every Real Estate Investor Needs to Know About Cash Flow" by Frank Gallinelli (Goodreads compounding) The Property Management Layer Determines Whether a Deal Performs as Underwritten "The Landlord's Survival Guide" by Janet Portman and Marcia Stewart makes the point that most value-add multifamily acquisitions underperform projections because the management execution wasn't scoped correctly. Turnover costs, maintenance response times and lease enforcement all compress net operating income in ways that didn't appear in the pro forma. Underwriting the management cost as a full line item, including the cost of switching operators mid-hold, changes what the realistic return looks like before you close. Source: "The Landlord's Survival Guide" by Janet Portman and Marcia Stewart (Goodreads compounding) --- SELF HELP, HUMAN PSYCHOLOGY & DARK PSYCHOLOGY Consistency Bias Locks You Into Positions You Should Abandon Robert Cialdini's broader body of work, specifically the commitment and consistency principle documented in "Influence" (already banned from direct citation, so this draws from "The Small BIG" by Steve Martin, Noah Goldstein and Robert Cialdini), shows that once people make a public commitment, their self-image becomes attached to being consistent with that position. The practical implication: decisions made publicly are harder to reverse even when the evidence shifts, because reversing feels like an identity violation rather than a rational update. The counter-move is to make provisional commitments explicit from the start, framing them as hypotheses rather than positions. Source: "The Small BIG" by Steve Martin, Noah Goldstein and Robert Cialdini (Goodreads compounding) The Spotlight Effect Makes You Overestimate How Much Others Notice Your Failures Thomas Gilovich's research, documented in "The Spotlight Effect" and covered extensively in "The Invisible Gorilla" by Gilovich and Christopher Chabris, demonstrates that people systematically overestimate how much attention others pay to their mistakes, appearance and performance. The cognitive cost of this bias is chronic self-monitoring that burns working memory and increases anxiety without improving actual outcomes. Knowing the spotlight is smaller than it feels is a practical license to act under conditions of imperfect preparation instead of waiting for certainty. Source: "The Invisible Gorilla" by Christopher Chabris and Daniel Simons (Goodreads compounding) --- WHAT TO WATCH The Storm-2945 device-code authentication pivot is the most operationally dangerous development this week. It converts Microsoft's own authentication infrastructure into the delivery mechanism, which means technical controls at the network or endpoint layer won't catch it. Watch for guidance from Microsoft on Conditional Access policy templates specifically targeting device-code flow, and watch whether other APT clusters adopt this technique in the next 30 days. --- CONVERSATION STARTER Storm-2945 drained MFA-protected M365 accounts by having victims authenticate on Microsoft's own sign-in page. The code came from the attacker, the session went to the attacker, and every security control in the stack saw a legitimate authentication event. The only mitigation is a Conditional Access policy that blocks device-code flow entirely before a traveler checks into a hotel. ===========================================