Russia's Camera Spy Network, CISA's Own Password Leak, and AI Hacking AI
Monday, July 20, 2026 · 5-minute read
Dutch intelligence agencies confirmed that Russian spies have been quietly hijacking IP cameras across Europe and Ukraine to watch military convoys, weapons deliveries, and troop movements. In Ukraine, the access went further — camera feeds were used to help target Ukrainian soldiers and destroy their equipment. Across NATO countries, the same operation is quietly gathering broader military intelligence.
The cameras are easy targets. Attackers scan the internet for devices with weak or default passwords and known software flaws. Many of these cameras sit in shops, on roadsides, and outside businesses — their owners have no idea they're streaming to Russian intelligence.
↗ The Hacker NewsMicrosoft's July Patch Tuesday was enormous: 570 vulnerabilities fixed, nearly triple last month's record. Microsoft says AI is helping its engineers find bugs faster — but that same speed benefits attackers too. Three of the flaws were zero-days, meaning hackers were already using them before the patch arrived. Two let attackers quietly gain higher-level control over a Windows machine.
One particularly eye-catching fix involves BitLocker — a bug that could let someone with physical access to your laptop read your encrypted data. If you use Windows, run updates now.
↗ Krebs on SecurityHugging Face, the platform where millions of developers share and download AI models, disclosed a breach this week with an unusual twist: the attacker was itself an autonomous AI agent. The agent exploited flaws in how the platform processes uploaded datasets to run its own code inside Hugging Face's systems. It accessed internal datasets and grabbed credentials used by Hugging Face's own services.
Hugging Face says no public-facing models or user data appear to have been tampered with, and its investigation is ongoing. But the incident shows that AI systems can now be weaponised to hack other AI systems — a feedback loop the security industry is only beginning to grapple with.
↗ BleepingComputerA critical vulnerability in the ServiceNow AI Platform — tracked as CVE-2026-6875 — is now being used in real attacks, according to threat intelligence firm Defused. ServiceNow is used by thousands of large enterprises to run their internal IT operations. The flaw allows an attacker to remotely run their own code on a vulnerable server without needing a password — the worst kind of bug. If your organisation uses ServiceNow and hasn't patched yet, attackers may already be looking for your instance.
Working exploit code for a set of critical remote code execution bugs in WordPress Core — nicknamed "wp2shell" — was released publicly this week. Once exploit code is public, automated scanners start hammering vulnerable sites within hours. Any WordPress site that hasn't applied the latest core update is a sitting target. An attacker who gets in can take over your site, steal visitor data, or turn it into a platform for spreading malware.
An unauthenticated attacker can send a specially crafted request to a ServiceNow server and run any code they want on it — no login required. This gives them full control of the server and everything connected to it. Attackers are already exploiting this in the wild.
Status: Patch available — apply immediately.
Opening a booby-trapped .xz archive in an older version of 7-Zip can let an attacker run code on your machine. The attacker doesn't get admin powers automatically — they inherit whatever rights 7-Zip already had — but it's still a real risk, especially if you regularly open files from untrusted sources.
Status: Fixed in 7-Zip 26.02, released June 25. Download the update from 7-zip.org.
SharePoint had a zero-day flaw that allowed attackers to gain elevated privileges on a Windows system — meaning they could quietly go from a normal user to someone with much wider access. This was one of three zero-days patched in Microsoft's July update, and it was already being exploited before the fix shipped.
Status: Patched in Microsoft's July 2026 Patch Tuesday. Install Windows updates now.
The US Cybersecurity and Infrastructure Security Agency (CISA) published a detailed advisory on July 13 urging individuals and organisations to clean up their router security — prompted directly by Russian state-sponsored groups scanning for weak home and business routers to use as anonymous relay points. The guidance is unusually practical for a government document: change default passwords, disable remote management if you don't need it, and keep firmware updated. The advisory came alongside a joint Dutch-US warning about the IP camera spy campaign. If you haven't touched your router's settings since you plugged it in, this advisory was written for you — the CISA page links to plain-English steps for major router brands.
↗ CISA Advisory AA26-194ATrend Micro researchers analysed 200 session logs from Google's open-source Gemini CLI tool and found a Russian-speaking criminal using it to manage a small botnet of hacked dental clinic computers, crack passwords, and plan cryptocurrency fraud targeting elderly people. The criminal used the AI to write attack scripts, migrate a command-and-control server, and set up anonymous browsing infrastructure. The entire operation fit in three tiny text files totalling 5KB — showing how low the barrier to entry for AI-assisted crime has become. This is the first well-documented public case of a criminal using an off-the-shelf AI coding tool to manage a live botnet.
↗ The Hacker NewsHere's an uncomfortable story: the US government body responsible for helping everyone else stay secure online — CISA — accidentally left a contractor's files full of its own passwords and cloud server keys in a public GitHub repository for nearly six months. A security researcher at GitGuardian spotted it. When they couldn't get CISA's attention through official channels, they eventually called in a reporter.
The leaked data included AWS GovCloud keys and a spreadsheet of plaintext usernames and passwords for dozens of internal CISA systems — stored in a file literally named "importantAWStokens." GitGuardian's automated systems had flagged the exposure nine times before the researcher escalated. CISA acknowledged the tip quickly but then took more than 48 hours to actually invalidate the leaked keys. The agency has since published an honest postmortem admitting its internal reporting channels were unclear and its response was slower than it should have been.
For regular people and organisations, the lessons are bigger than just "CISA messed up." First: even security experts make basic mistakes — leaving credentials in files, not rotating keys, ignoring automated alerts. Second: when someone tries to report a problem to you, make it easy for them. CISA's own postmortem says the researcher had to try emailing the contractor, submitting through a vulnerability form, and calling a journalist before anyone acted. That's a process failure, not a technical one. Third: the gap between "keys were leaked" and "keys were revoked" is a window attackers love. In this case it was six months wide.
CISA is now refining how it receives security reports about its own infrastructure, separate from reports about the wider internet. Watch for whether the agency updates its public security.txt file and reporting instructions — that will be the real signal that lessons were absorbed, not just documented.