Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

NewsIncident

Unit 42: model-written scripts, a Go backdoor and a fake sudo prompt in a six-month espionage campaign against government Linux servers

Palo Alto Networks Unit 42 described CL-UNK-1211 on October 15, with coverage on October 17: a campaign from October 2025 to April 2026 against government entities and state-affiliated infrastructure on Linux and database servers, countries unnamed. The attackers' PHP and shell scripts carried the conversational annotations and step-by-step reasoning that models write, switching between PHP, JavaScript, Bash and Python, which Unit 42 reads as model-generated or agent-run code without proof of autonomy; one script would have failed on a syntax error. They kept access with web shells and by making every PHP configuration load their code first, deployed SysGoRAT, a Go backdoor for Linux and Windows over WebSockets, and replaced .bashrc with a fake sudo prompt that logged passwords to a file named after SolarWinds, which was not compromised. They took administrator logins from modified login pages, mail and database passwords from configuration files, mailboxes with social media password resets, and database tables dumped to CSV, then cleared logs, backdated files, and planted a "RedTeamOps2025" key and a "pentest123" password to pass as an authorized test. Unit 42 has not decided whether the actor is a state or a criminal one.

Why it matters

Model-written tooling is ordinary in intrusions now, and it is recognizable by its comments. Restrict what your servers can reach on the internet, alert when a utility spawns a shell, and treat a red-team marker you did not plant as an intrusion rather than a reason to stand down.

Sources

Read it with the rest of the news in the Learning Center