AI Hacked OpenAI, North Korea Stole $10M, and a 9.8 Flaw Is Being Exploited Now
Saturday, September 19, 2026 · 5-minute read
A joint law enforcement advisory confirmed this week that the North Korean hacking group WaterPlum compromised at least 30,000 devices worldwide between December 2025 and July 2026. The group then funnelled more than $10.7 million in stolen cryptocurrency back to North Korea. This is a reminder that state-sponsored hackers aren't just spying — they're actively stealing money to fund the regime.
↗ BleepingComputerThe ShinyHunters extortion gang broke into the Clop ransomware operation's own data leak website. They defaced it, allegedly stole server data, and grabbed the private keys for Clop's hidden Tor site. Criminals attacking other criminals is unusual — and it could mean stolen data or disrupted operations for one of the world's most active ransomware groups.
↗ BleepingComputerA New Jersey judge ordered the domain radaris.com — and more than a dozen related people-search sites — transferred to the plaintiffs suing the company. Radaris had spent years dodging accountability by cycling through shell companies in places like the Marshall Islands and the British Virgin Islands. The lawsuit was brought under New Jersey's Daniels Law, which lets law enforcement officers demand their personal info be removed from data broker sites.
↗ Krebs on SecurityAttackers are actively exploiting a critical flaw in Orkes Conductor, a platform businesses use to build and run automated workflows. The bug — rated 9.8 out of 10 in severity — lets anyone on the internet run commands on a vulnerable server without needing a password or account. The attacker simply sends a specially crafted request to the workflow API, and the server obeys. Fortinet confirmed active exploitation this week, meaning real attackers are using this right now, not just researchers.
Criminals are creating fake GitHub repositories that impersonate well-known software brands — including LastPass — to trick people into downloading a new infostealer called Rapuncel. They're using search engine tricks to push these fake pages to the top of Google results. If you download what looks like a legitimate app but is actually Rapuncel, it quietly harvests your saved passwords and sends them to the attackers.
This flaw lets a complete stranger run any command they want on a server running Orkes Conductor — no login required. They do it by slipping malicious JavaScript or Python code into a workflow definition and sending it to an open internet connection point. The server runs the code without questioning where it came from.
Status: Patch available — upgrade to Orkes Conductor 3.30.2. Actively exploited in the wild.
SolarWinds left a hard-coded static key inside its Access Rights Manager software. Anyone who finds that key can use it to remotely run code on an affected system without ever logging in. Think of it like a master key that was accidentally printed on the outside of every lock. All versions up to ARM 2026.2 are affected.
Status: Patch available — update to ARM 2026.2.1. No known exploitation in the wild yet.
Cisco is warning of a maximum-severity zero-day flaw in its Identity Services Engine, a product companies use to control who gets access to their network. Cisco says this vulnerability is already being exploited in real attacks. A successful attack could give an outsider full control over the system that's supposed to be keeping outsiders out.
Status: Actively exploited. Check Cisco's security advisories immediately for available patches or workarounds.
Gyazo, a popular tool that lets you instantly share screenshots online, confirmed hackers exploited a server flaw to steal 23.6 million user records. The breach is a reminder that even simple utility apps you barely think about hold real personal data. If you have a Gyazo account, change your password now and watch for phishing emails that might use your account details to seem more convincing. This also highlights a broader pattern: attackers increasingly go after smaller, less-defended platforms that people use casually.
Microsoft is rolling out a new Teams feature that lets company administrators customize which file types are automatically blocked from being shared in chats. Right now, Teams blocks a fixed list of dangerous file extensions. Soon, IT teams will be able to add their own extensions to that list or fine-tune it for their specific needs. This matters because attackers frequently use Teams to deliver malware disguised as ordinary documents — especially in business email compromise-style scams that have moved to chat platforms.
Three researchers at a security firm called Hacktron used Anthropic's Claude Opus 5 AI to find and chain together two separate security flaws — and within 72 hours, they had access to the internal code repository of OpenAI, the company behind ChatGPT. This wasn't a criminal attack. It was authorized security research. But the implications are genuinely significant: an AI was used to speed up the process of hacking another AI company.
Here's how the chain worked. The researchers first found a bug in the software running OpenAI's public help forum. That forum runs on a widely used open-source platform called Discourse. That bug alone wasn't enough, but it opened a door. They then found a weakness in OpenAI's own login system. By combining the two flaws — a technique called vulnerability chaining — they were able to take over actual employee accounts and reach internal code. To prove they had access without causing damage, they made a harmless pull request (basically a minor code suggestion) and stopped. They then reported everything to OpenAI.
OpenAI's response was fast. The company confirmed a fix within 14 hours of receiving the report and paid the team a $6,500 bug bounty. OpenAI noted that the bounty covered the flaw in its own login system — not the forum software bug, which was outside the scope of its bounty program. The company hasn't publicly described exactly what the login flaw was.
What makes this story worth sitting with is the AI angle. Using Claude Opus 5 to help discover and chain vulnerabilities compressed what might have taken a skilled human team days or weeks into a much shorter window. Security researchers are now openly using AI to find flaws faster — but so are criminals. The gap between "a vulnerability is discovered" and "attackers are exploiting it" is shrinking. For most of us, the takeaway is simple: the companies behind the tools you use every day need to patch faster than ever, and the pressure on them to do so is growing by the week.