📰 The CyberSignal Daily — August 4, 2026

First, a quick note: The CyberSignal Daily is back on a regular cadence — a tightly-edited morning briefing in your inbox every day. Thanks for sticking with us through the pause. Now, on to it.

Yesterday's theme was autonomy — and how little of it anyone fully controls. Unit 42 detailed a Chinese-speaking operator who ran DeepSeek through the Hermes agent framework to attack a security firm and queue 1,200+ hosts for proxyjacking, until its own AI agent exposed the whole operation. Unit 42 also showed how ordinary Windows malware can hijack Google passkeys with no fingerprint, the US water campaign was confirmed across seven states as President Trump publicly rejected his own agencies' Iran assessment, and the N-able N-central bypass got worse when its first patch was itself bypassed. Our new explainer ties the AI thread together.

AI Security · Threat Actors

Unit 42's full writeup details a Chinese-speaking operator (tracked as knaithe / KnYuan) who wired DeepSeek into the Hermes agent framework to run attacks with limited human intervention — targeting a security firm and queuing more than 1,200 hosts for proxyjacking and follow-on attacks. What makes the case remarkable is how it unraveled: the operation's own AI agent misconfigured a file server and published the entire back end — API keys, exploit scripts, target lists — to anyone who looked.

It's the clearest picture yet of autonomous offense: real, scalable, and prone to machine-speed mistakes. Defenders should tune for machine-cadence enumeration — fast, consistent, tireless — and the tells of proxyjacking, like unexpected residential-proxy traffic and bandwidth that doesn't match a host's job.

→ Read the full story

Critical Infrastructure

The water-sector campaign now spans Minnesota plus at least six more states, with Michigan, Georgia, and South Dakota named; a treatment plant in Braham was briefly disrupted at the industrial-control layer. Officials describe the activity as likely tied to Iran — but President Trump publicly rejected that read and blamed Minnesota's "grossly incompetent" governor, putting him at odds with his own agencies.

The attribution fight is loud, but it doesn't change the checklist. Whoever is behind it, water utilities should audit internet-exposed PLCs, segment or disable remote-management interfaces, rotate credentials on internet-facing OT, and rehearse manual-operations fallback.

→ Read the full story

Cloud & Identity

Unit 42 detailed three attack paths — Pass-ta-key, Silver, and Golden — against Chrome's Google Password Manager acting as a cloud authenticator. Running as an ordinary Windows user, malware can produce a valid passkey sign-in with no fingerprint, PIN, or on-screen prompt; the strongest path extracts the 32-byte master secret that decrypts every synced passkey you own. The cryptography isn't broken — the operational plumbing around it is.

Passkeys are still far better than reused passwords, but "we use passkeys" stopped being a complete answer. Prefer device-bound passkeys — hardware keys, a Windows Hello TPM, on-device Apple keychains — for crown-jewel accounts, and if you run a relying party, set userVerification to required and actually validate the User Verified flag.

→ Read the full story

Vulnerabilities · Supply Chain

Our full breakdown of CVE-2026-18577: an actively exploited authentication bypass in N-able's N-central RMM whose first fix was defeated by attackers. Build 2026.3.1.7, released August 2, is the first unaffected version. Because N-central sits atop the MSP management chain, one compromised server can reach every managed endpoint beneath it — the same structural weakness behind the Kaseya and SolarWinds incidents.

If you run N-central on-premises, treat it as an active incident: upgrade to 2026.3.1.7, audit administrative logs back to at least July 31, assume downstream reach until you can rule it out, and notify affected customers.

→ Read the full story

Cybersecurity 101

Yesterday's autonomous-agent and passkey stories share a root cause: machine-learning systems can be attacked through the inputs they process, the queries they answer, and the data they trained on. Our new primer walks the major attack categories — evasion, model extraction, membership inference — explains why ML is vulnerable by design, and lays out the layered defenses that actually help.

The takeaway for anyone shipping ML: treat these systems as high-value production assets, not experiments — evaluate them, red-team them against adversarial inputs, and monitor them in production.

→ Read the full story

Markets don't wait for Monday. News breaks on a Saturday morning, and most traders can do nothing but watch.

Not on Liquid. Trade domestic and international equities, commodities, forex, crypto, and prediction markets — all from one account, 24 hours a day, 365 days a year. Liquid gives you access to any market, from anywhere, anytime. To us, access is arbitrage.

Getting started takes under 10 minutes: log in with Google, deposit with Apple Pay or a bank transfer, and trade from your phone or desktop — wherever you are in the world.

While everyone else is refreshing headlines and waiting for the open, you're already positioned. That's the difference between reacting to markets and actually trading them.

Trade 24/7 on Liquid Today

The throughline yesterday was autonomy — and how little of it is under anyone's full control. A threat actor handed a campaign to an AI agent and mostly stepped back; the agent reached a security firm and lined up more than a thousand hosts before tripping over its own misconfiguration. That's the shape of offense now: fast, broad, and sloppy in machine-specific ways. The sloppiness is, for the moment, one of the better gifts defenders are going to get.

The trust stories cut the other way. Passkeys were sold as the end of the password era, yet Unit 42 showed commodity malware reaching the synced key store without ever touching the cryptography. And our adversarial-ML primer explains why the layer underneath all of this — the models themselves — is attackable by design, not by bug. "We use passkeys" and "the model is accurate" are both incomplete answers now.

Then there's the unglamorous half no model can paper over: an RMM patch that got bypassed, and an OT campaign creeping across state water systems while the politics get louder. The synthesis for defenders is the same as it's been all week — assume autonomy on offense, stop treating any single control as finished, and keep the management-plane and OT fundamentals tight, because that's where one compromise turns into many.

Stay sharp. Stay ahead.

Till next time,

The CyberSignal Team

Our News Site | Get the Weekly Briefing