AI and IT News Recap: August 5, 2026: A Self-Spreading npm Worm Poisons 400+ Packages, Russian Spies Hijack Hotel Wi-Fi, and Microsoft Open-Sources the Agent Training Layer

By Noah Smith, Owner & Consultant, KeyChange Technologies · August 5, 2026

If you run anything that touches npm, or your team travels for work, today's AI and IT news is aimed squarely at you. The last 24 hours brought a self-spreading supply-chain worm with a genuinely huge blast radius, a Russian espionage campaign that turns hotel Wi-Fi against business travelers, and a quieter but important move from Microsoft to open up how AI agents get trained. Picking up from yesterday's recap, here is what actually matters.

📌 The AI and IT news at a glance

• 🪱 A self-spreading npm worm poisoned 400+ packages with over 2 billion monthly installs, starting from the popular keyv library.
• 🏨 Russian spies (Midnight Blizzard) hijacked hotel Wi-Fi to steal Microsoft 365 tokens with two custom malware strains.
• 🔓 An actively exploited N-able N-central flaw let attackers walk into the tools your IT provider uses to manage your machines.
• 📡 Researchers found 15 flaws in TP-Link Omada gear that can chain into a full network takeover.
• 🎣 A 89-a-month phishing kit added a slicker way to beat MFA by abusing legitimate login codes.
• 🤖 Microsoft open-sourced Orchard, a framework that makes training capable AI agents far cheaper.
• 🌐 Anthropic hired a former California Supreme Court justice as its first Chief Global Affairs Officer.

🔝 Top story: a self-spreading worm turned trusted npm packages into a credential heist

The biggest story of the day is a supply-chain attack that spreads on its own. Attackers compromised the GitHub account of the maintainer behind keyv, a small but wildly popular key-value storage library with roughly 127 million weekly downloads, and used that access to plant a credential-stealing worm across the entire package family. It did not stay put. By the time researchers finished counting, at least 434 packages across roughly 1,381 versions had been poisoned, with a combined total of more than 2 billion monthly installs. Security firm Aikido ties the activity to the "Shai-Hulud" worm family.

The mechanics are what make this nasty. Every poisoned release quietly adds a single line, a "preinstall" script, so the malicious payload runs the moment someone types npm install, before any of their own code executes. Once running, it hunts for GitHub, npm, AWS, GCP, Azure, Stripe, Vault, and Kubernetes credentials, plus database connection strings and private keys, then uses any npm publishing access it finds to poison still more packages and keep the chain going. Some variants even planted hooks into Claude Code and VS Code, meaning a developer's own AI tooling could become part of the spread.

In short: A worm hidden in updates to the widely used keyv npm library stole developer and cloud credentials and self-propagated to more than 400 packages with billions of monthly installs.

What it means for your business: If your product, website, or internal tools are built on JavaScript, your developers almost certainly pull from npm, and one careless npm install this week could have handed over the keys to your cloud accounts. Freeze dependency updates, pin versions, rotate any credentials that touched a build machine in the last several days, and ask whoever owns your codebase whether they audited recent installs.

My take: The scariest part is not the number of packages, it is the word "preinstall." You do not have to run the bad code or import it anywhere; simply installing sets it off. That breaks the mental model a lot of teams still have, where they assume they are safe until they actually call a library. Treat your build pipeline like production, because attackers already do.

Source: The Hacker News

🛡️ IT and security

🏨 Russian spies are hijacking hotel Wi-Fi to raid Microsoft 365 accounts

Microsoft says the Russian state-sponsored group Midnight Blizzard (also known as APT29), working through a sub-team it tracks as Storm-2945, has been tampering with the captive-portal networks at hotels, conference centers, and shared venues to redirect guests toward malware. Microsoft calls the campaign CaptiveCrunch and believes it has been running since at least May 2026. When a traveler connects and hits the usual "sign in or accept terms" page, manipulated DNS and HTTP traffic quietly steers their device to a fake update or login instead.

Two custom malware strains do the dirty work. CornFlake is a Go-based remote access trojan that hides behind a fake progress window, installs itself as a Windows service disguised as "Cloud Sync Service," and then logs keystrokes, grabs screenshots, and taps the microphone and webcam. ChocoShell is a PowerShell infostealer that runs entirely in memory, switches off Windows' built-in script scanning, and harvests browser passwords, cookies, Microsoft 365 single sign-on tokens, and saved Wi-Fi credentials.

In short: A Russian espionage group is abusing hotel and venue Wi-Fi to push malware that steals Microsoft 365 login tokens from traveling employees.

What it means for your business: Any employee who logs into work accounts from a hotel network is a potential entry point, and stolen session tokens can sidestep multi-factor authentication entirely. Push staff to use a VPN or a phone hotspot on the road, be suspicious of "update required" prompts on public Wi-Fi, and consider requiring re-authentication for sensitive systems after travel.

My take: This is a good reminder that "we have MFA" is not the finish line. Token theft is the current fashion precisely because it skips the password and the second factor together. The practical defense here is boring but effective: stop trusting captive portals, and get people off hotel Wi-Fi for anything that matters.

Source: BleepingComputer

📡 A live N-able N-central flaw puts managed IT tools at risk

CISA has added CVE-2026-18577, an actively exploited authentication-bypass flaw in N-able N-central, to its Known Exploited Vulnerabilities catalog. N-central is a remote monitoring and management platform that managed service providers use to run and support their clients' computers, which makes it a high-value target: break the platform, and you can reach every managed endpoint behind it. N-able says exploitation has been seen in the wild since August 1, and that attackers used N-central's own "Take Control" remote-access feature and set up Cloudflare Tunnel to keep a foothold. Awkwardly, the flaw exists because an earlier patch for a related bug (CVE-2026-18556) was incomplete.

In short: Attackers are actively exploiting an authentication-bypass bug in N-able N-central, the software many IT providers use to manage customer machines, to gain admin access.

What it means for your business: If you outsource IT, your provider very likely uses a tool like this, and a compromise there flows straight down to you. Ask your MSP directly whether they run N-central and whether they have applied the 2026.3.1.7 hotfix. If you run it yourself, patch now.

My take: Supply-chain risk is not just the software you buy, it is the software your vendors use to touch your systems. This one deserves a two-line email to your IT provider today. Their answer tells you a lot about how quickly they move.

Source: BleepingComputer

📶 Fifteen new TP-Link Omada flaws can chain into a full network takeover

Researchers at Forescout's Vedere Labs disclosed 15 vulnerabilities in TP-Link's Omada networking ecosystem, the switches, access points, and controllers common in small and mid-size office networks. The findings include hardcoded cryptographic keys, credentials sent insecurely, weak certificate checks that enable man-in-the-middle attacks, and a cross-site scripting bug in the controller's web interface. Eleven of the 15 received CVE identifiers; TP-Link declined to assign the other four, citing low severity. Chained with two older remote-code-execution flaws, the researchers demonstrated realistic paths to compromise whole fleets of managed devices, and they found roughly 1,800 Omada controllers exposed directly to the internet.

In short: Forescout found 15 flaws in widely used TP-Link Omada network gear that can be combined to take over entire managed networks, with patches only partially available.

What it means for your business: If your office runs TP-Link Omada equipment, get your Omada controller off the public internet immediately and apply available updates. Some structural fixes will not land until later in 2026, so network segmentation and restricted management access matter here.

My take: Networking gear is the classic "set it and forget it" purchase, which is exactly why attackers love it. The single most useful action from this story costs nothing: make sure your management console is not reachable from the open web.

Source: SecurityWeek

🎣 A cheap phishing kit gets better at beating MFA

The commercial phishing-as-a-service toolkit known as Greatness added support for "device code phishing," a technique that abuses the legitimate OAuth 2.0 device-authorization flow (the "enter this code on another device" pattern) to trick users into approving an attacker's login and hand over session tokens without a password. Greatness bundles this with adversary-in-the-middle token theft and OAuth consent abuse in one operator panel, targets Microsoft 365, iCloud, Yahoo, and Google Workspace, and sells for about 89 a month over a Telegram channel with thousands of subscribers.

In short: A 89-a-month phishing kit now offers device-code phishing, a growing method that beats multi-factor authentication by tricking users into approving a login code.

What it means for your business: The barrier to running a token-stealing phishing campaign against your staff is now a monthly subscription. Move toward phishing-resistant MFA (passkeys or hardware keys), disable device-code authentication where you do not need it, and specifically train employees to distrust any unexpected "enter this code" request.

My take: When advanced attacker tradecraft shows up in a 89 subscription, it is no longer a nation-state problem, it is an everyone problem. The good news is that this specific trick has a clean fix: turn off the device-code flow if your organization does not use it.

Source: The Hacker News

🤖 AI and agents

🌱 Microsoft open-sourced Orchard, the missing training layer for AI agents

Microsoft Research released Orchard, an open-source framework for training and evaluating AI agents, and analysts are calling it a missing layer in the agent stack. Most open tooling so far has focused on how agents run, not how they learn; Orchard tackles the harder, more expensive part, providing a lightweight Kubernetes-based environment (Orchard Env) with reusable, isolated components for collecting training data, running reinforcement-learning rollouts, and evaluating results. It ships with three recipes: Orchard-SWE for software engineering, Orchard-GUI for browser navigation, and Orchard-Claw for everyday productivity tasks. The headline result is efficiency: Orchard-SWE reaches 69.7% on the SWE-bench Verified coding benchmark (73.0% with reranking) using only about 3 billion active parameters, approaching systems more than 10 times larger.

In short: Microsoft open-sourced Orchard, a framework that makes it cheaper to train capable AI agents, with a small model nearly matching far larger ones on a coding benchmark.

What it means for your business: This is infrastructure, not a product you will buy tomorrow, but it points to where things are heading: smaller, cheaper, task-specific agents that a mid-size company can actually afford to run. Over the next year, expect more capable automation at lower cost per task.

My take: The interesting number here is not the benchmark, it is the parameter count. A 3-billion-parameter model getting near frontier coding scores means good agents are getting small enough to run cheaply and privately. That trend matters far more to most businesses than which lab has the biggest model this month.

Source: Microsoft Research

🌐 Anthropic names its first Chief Global Affairs Officer

Anthropic hired Mariano-Florentino (Tino) Cuellar as its first Chief Global Affairs Officer, a role focused on policy, international engagement, and government relationships as AI regulation heats up worldwide. Cuellar recently stepped down as president of the Carnegie Endowment for International Peace and was previously a Justice of the Supreme Court of California, where his opinions touched on technology, privacy, and the separation of powers. He will report to Anthropic president Daniela Amodei and manage the company's relationships with the U.S. government and the many countries where it is expanding.

In short: Anthropic appointed a former California Supreme Court justice and Carnegie Endowment president as its first global policy chief.

What it means for your business: Leadership hires like this signal that AI rules, licensing, and cross-border data questions are becoming central to how these vendors operate, which eventually shapes the contracts, compliance terms, and availability of the AI tools you use. It is a background trend worth tracking, not an action item.

My take: When a frontier lab recruits a former state supreme court justice to run global affairs, it is telling you the next phase of this industry will be fought in policy and courtrooms as much as in model quality. For customers, that mostly means more stability and clearer rules over time, which is not a bad thing.

Source: Anthropic

That's the AI and IT news for August 5, 2026. If you only do one thing today, check with whoever manages your code or your IT provider about the npm worm and the N-able flaw.