
What happened
Dream researchers found artifacts of a campaign that rooted 25 PAN-OS firewall and management servers across the US, Japan, Taiwan and South Korea, including three in a US government-affiliated environment.
Why it matters
The attacker turned an exposed management interface into a path into identity infrastructure, and 22 of the rooted servers were left parked with active webshells, so they remain exploitable.
WHO IT HITSGovernment and critical-infrastructure network defenders running exposed PAN-OS management interfaces are the most directly affected; the artifacts show the attacker reached Active Directory from a compromised management server, so identity teams may need to assess credential exposure.
Summaries like this, every morning.
The operation stands out less for a novel vulnerability than for the preparation behind it. According to Dream, the attacker obtained PAN-OS firmware, decrypted it, extracted the PHP libraries and built 11 JSON version profile files, one for each PAN-OS version tested, complete with offsets, SHA256 hashes and memory geometry parameters tuned through lab testing. That investment let the exploitation itself run as an automated batch against 240 targets, five at a time with a 30-minute timeout per target.
The targeting funnel was equally deliberate. FOFA surfaced roughly 19,000 exposed management interfaces; a 200-thread scanner cut that to about 5,000 live devices reporting version information; geographic and version filtering reduced it to 240. Dream notes that the four countries selected are the primary strategic adversaries and intelligence targets for Chinese state-aligned operations in the Pacific theater, which it says indicates potential nation-state adversary efforts, while adding that it does not attribute the activity to a named threat actor. Palo Alto Networks' Unit 42 tracks it as CL-UNK-1280 and attributes it to a Chinese-speaking attacker in the UTC+8 time zone.
What Dream could not recover is the result of the later stages. The password spray against every domain user, the attempts to create backdoor user and machine accounts, and the SAML signing key extraction all ran from the management server, but their outputs stayed on that server's local filesystem and were not among the artifacts. Dream states plainly that it cannot determine the full extent of the compromise, and that the persistent tunnel and root webshells are likely still active.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
OpenAI restructured its public Model Spec around five core values: faithfulness to human intent, harm reductio…

A-RSI, published on arXiv on October 28, 2026, refreshes a model's task pool when advantage retention drops be…

Tyler Shultz, the Theranos whistleblower, told The San Francisco Standard that AI is 'a great opportunity to g…

OpenAI says it made over 200 law enforcement referrals in Q3 2026 and will publish a Community Safeguards Repo…

Anthropic's prospectus devotes 69,800 words, about 30% of its 240,000-word filing, to risks like job displacem…

OpenAI CEO Sam Altman said on the Cheeky Pint podcast that humanity “deserves fundamental knowledge” and that…
