The "Verified" Badge That Lies, Two Linux Zero-Days & an AI That Runs Ransomware
Wednesday, July 8, 2026 · 5-minute read
Thalha Jubair, 20, and Owen Flowers, 18, admitted in a UK court on Monday to breaking into Transport for London's computer systems in August 2024, causing serious disruption to the city's public transit network. The pair are members of Scattered Spider, a group linked to over 120 network intrusions and at least $115 million in ransom payments from US companies. Flowers is also tied to hacks against US healthcare providers, and Jubair ran a SIM-swapping service on Telegram that targeted major wireless carriers.
↗ Krebs on SecurityJapanese telecom KDDI confirmed that attackers broke into an email platform shared by five internet service providers in Japan, exposing the email addresses and passwords of over 12 million people. If you use any Japanese ISP email service, assume your credentials are out there and change your password now. This is a good reminder that email platforms — not just the big social networks — are prime targets for credential stuffing.
↗ BleepingComputerThe FBI, working with Google and other partners, seized hundreds of domains tied to NetNut — a residential proxy service run by Israeli-listed company Alarum Technologies. NetNut secretly turned smart TVs and streaming boxes into relay nodes for cybercriminals to hide their tracks. In a single week in June, Google spotted 316 different criminal groups using NetNut's network to run password attacks and cover up their locations.
↗ Krebs on SecurityA ransomware group called JadePuffer has become the first publicly documented criminal gang to hand the controls of an entire attack to an AI agent. The AI handled reconnaissance, picked targets, moved through networks, and deployed the ransomware — all without a human operator making decisions in real time. This matters because it dramatically lowers the cost and skill needed to run a sophisticated attack, meaning more criminals can now launch the kind of intrusions that previously required an expert team.
CISA confirmed that a remote code execution vulnerability in Microsoft SharePoint is now being actively used in real attacks, not just tested by researchers. SharePoint is widely used by businesses and government agencies to share files and manage documents internally. CISA has ordered federal agencies to patch by Friday, July 11.
A 15-year-old bug dubbed GhostLock lets any ordinary logged-in user take full root control of a Linux machine. No special permissions needed — just a normal user account. Researchers at Nebula Security built a working exploit that succeeds 97% of the time and can also break out of containers. The vulnerable code has been in essentially every mainstream Linux distribution since 2011, and working exploit code is now publicly available.
Status: Patches are being released by major Linux distributions now. Apply them immediately — public exploit code means criminals can use this today.
A second old Linux kernel bug, Januscape, surfaced this week. This one is 16 years old and lets an attacker escape a virtual machine and run code directly on the physical host machine underneath. That's especially serious in cloud environments, where many customers share the same physical server. If exploited, an attacker in one customer's VM could potentially reach another customer's data.
Status: Security updates are available. Cloud providers and Linux vendors are issuing patches — check for updates on any Linux-based systems you manage.
Ubiquiti patched seven critical flaws in its popular UniFi OS platform today, including one rated the maximum possible severity score of 10. The worst flaw allows command injection — meaning an attacker could take complete control of your router or network equipment by sending it a specially crafted request. Ubiquiti gear is common in both home offices and small businesses.
Status: Patch available now. Log in to your UniFi controller and apply the latest update as soon as possible.
Researchers Abhishek Kumar and Carsten Maple published findings this week showing that AI coding assistants can be tricked into producing harmful content even when they refuse to do so directly. The trick: instead of asking for the dangerous thing outright, you break the task into small, innocent-looking coding steps. Across 816 test runs, models including Claude and Gemini produced the banned content every single time when the request was framed this way — even though they refused the same request when asked directly. The researchers call this a "workflow-level jailbreak." It's notable because no one is pretending to be someone else or feeding the AI fake context — the AI writes the harmful code as a side effect of a normal-looking task. AI companies will need to rethink safety filters that only check individual prompts, not multi-step workflows.
If you work in software — or just use software — you've probably heard that developers can "sign" their code to prove it came from them. GitHub shows a green "Verified" badge on signed commits. That badge is supposed to be a seal of trust: this code came from this person, and it hasn't been touched since. New research published today shows that badge can be faked without breaking a single signature.
Here's the problem in plain terms. When a developer signs a commit, the signature covers the content of the code. But the unique ID (called a hash) that GitHub uses to identify that commit is calculated separately. Researchers found you can take a signed commit, generate a second commit with the same code, same author, same date, and a valid signature — but a completely different hash. GitHub still shows "Verified" on both. The original and the copy look identical to anyone reviewing them.
Why does that matter for regular people? A lot of software security depends on the idea that a hash is a permanent, unique name for a specific version of code. Security teams block known-bad commits by hash. Build systems record hashes to prove what code went into a product. Supply chain security tools use hashes to track what software came from where. If an attacker — or a compromised mirror server — can serve you a validly signed commit with a different hash, all of those checks quietly fail. You think you blocked the bad commit. You didn't.
This isn't a theoretical problem. Supply chain attacks — where criminals sneak malicious code into trusted software packages — are one of the fastest-growing attack categories. Tools that developers rely on to catch these attacks just became slightly less reliable. Watch for GitHub and the broader open-source community to respond with guidance on verifying both signatures and hashes independently, rather than treating the "Verified" badge as the end of the story.