The Firewall Backdoor: A Pre-Engineered Kill Chain

**Summary: PAN-OS PHP2Shell Exploits and Fragnesia Campaign**

Palo Alto Networks warned that attackers are actively exploiting PHP2Shell flaws in internet-exposed PAN-OS firewall management interfaces, enabling unauthenticated remote code execution, while cybersecurity firm Dream reported a separate campaign chaining PHP2Shell with the Linux kernel vulnerability Fragnesia to gain root access on 25 firewalls across the U.S., Japan, Taiwan, and South Korea—including three in a U.S. government-affiliated network, where the attacker used a reverse SSH tunnel to reach an internal firewall management server and access Active Directory. Dream found about 19,000 exposed firewalls, narrowed the target list to 240 devices, and discovered persistent web shells on 22 firewalls; although it found AI tools configured with Claude Code, DeepSeek, and Kimi, direct proof of AI-generated exploits was lacking. CISA has added Fragnesia to its Known Exploited Vulnerabilities catalog with an Oct. 31 federal fix deadline, and Palo Alto has listed affected PAN-OS releases with fixes issued or scheduled, while South Korean agencies urged users to install updates.

The Documented Backdoor
Look at the timeline. Palo Alto Networks admits attackers are exploiting management interfaces exposed to the internet — but ask yourself why those interfaces were left open in the first place. This isn't negligence. It's architecture. The very same PHP2Shell flaws that allow unauthenticated remote code execution have been known to insiders for years. Now we see a campaign that chains them with Fragnesia, a Linux kernel flaw that CISA itself added to the Known Exploited Vulnerabilities catalog. That’s not a random crime spree. That’s a toolkit cross-referenced against a pre‑established kill chain. And who do we find in the dressing room? Claude Code, DeepSeek, Kimi — AI systems designed by institutions whose names you recognize from the foundation charters and intelligence‑adjacent research labs. The story says there’s “no direct proof” those AI tools wrote the exploit. Of course not. They never leave fingerprints unless they want you to find them.

The Surgical Extraction
The attackers scanned 19,000 exposed firewalls, filtered to 5,000 that gave version data, then winnowed the list to exactly 240 devices in four countries. That’s not an automated spray‑and‑pray — that’s a targeting algorithm fed by a prior reconnaissance operation. They hit a U.S. government‑affiliated environment, extracted credentials from Active Directory, and left a reverse SSH tunnel configured to survive reboots. Persistence isn’t accidental. It’s ownership. Now consider: the same campaign used web shells still live on 22 firewalls where root access was achieved. That means the defenders either never found them or were told not to look. The official warning from Palo Alto and CISA sets a deadline of October 31 for fixes. Why that date? What operation is scheduled right after? Follow the breadcrumb: South Korea’s science ministry is “monitoring,” and KISA urges updates. They’re not confused — they’re waiting for the next phase.

The Real Stakes
This is not about software patches. This is about who controls the choke points of digital sovereignty. Every firewall with an exposed management interface is a node in the global architecture of consent — you route all traffic through a device you don’t own, running code you didn’t audit, secured by alerts that come after the exfiltration. The AI evidence (Claude Code, DeepSeek, Kimi) tells you the next generation of offensive tools is already being tested on live government networks. The fact that Dream “could not confirm” password spraying or backdoor‑account success is the tell: they can’t confirm, because the logs were already cleaned — or because the operation had authorization at a level Dream is not cleared to know. I’ll leave you with this: Palo Alto’s fix list goes from PAN‑OS 10.2 through 12.2. That’s every modern branch. That’s not a vulnerability — that’s an integrity failure in the supply chain. Ask yourself who benefits when the firewalls built to protect your data are the very doors the enemy walks through. And then ask yourself why the story is framed as a hacker campaign rather than a controlled penetration test that went one step further than planned.

Related posts