Hackers breached 25 servers, including US government-linked systems; Claude Code and other AI tools were found among their tools
Dream Research reported that attackers exploited PHP2Shell and Fragnesia flaws to breach 25 servers (24 firewalls and an internal management server) in the United States, Japan, Taiwan and South Korea, after scanning roughly 19,000 internet-facing firewalls. At a U.S. government-affiliated organisation they compromised three servers, stole an account password and obtained user and group lists. Researchers found Anthropic's Claude Code and an AI-powered platform called aiscan configured with DeepSeek and Kimi in the attackers' environment, plus about 40 scripts that may have been AI-written. However, Dream Research cautioned there is no direct evidence AI was used to create the exploits or carry out the attacks.
What the AI did
Anthropic's Claude Code and an AI-powered security platform called aiscan, set up to work with the Chinese AI models DeepSeek and Kimi, were found among the attackers' tools, along with about 40 scripts that may have been written with AI help. Researchers said there is no direct evidence that AI was used to create the exploits or carry out the attacks.
First reported October 28, 2026 · Added to the register November 5, 2026 · 1 source
- Developer
- Anthropic
- Model
- Claude Code; aiscan, DeepSeek and Kimi
- Affected
- U.S. government-affiliated organization
- Country
- Japan, South Korea, United States, Taiwan
- When it happened
- Not stated in the sources
- First reported
- October 28, 2026
What this means for you
Could this affect you?
If you run internet-facing firewalls or PHP-based web servers. Organisations running internet-facing firewalls and PHP-based web servers in the US, Japan, Taiwan and South Korea were exposed, including a U.S. government-affiliated organisation whose account password and user and group lists were taken.
What to check
- Patch PHP against the PHP2Shell flaws.
- Patch the Fragnesia Linux vulnerability.
- Check for web shells and other signs of compromise, because patching will not remove access attackers already have.
- Monitor for attackers misusing AI coding and scanning tools.
Areas of your AI programme this touches
Every fact and its source (8)
- ResponseFirewall vendor notified“says it has notified the vendor”[1]
- Data stolenAccount password from central firewall management server“compromising a central firewall management server and stealing an account password”[1]
- Scanning scaleAbout 19,000 firewalls scanned, 240 targeted“scanning roughly 19,000 internet-facing firewalls before narrowing their targets to 240 potentially vulnerable systems”[1]
- Breach breakdown24 firewalls plus one internal management server“they used automated attacks to breach 24 firewalls and an additional internal management server, bringing the total to 25”[1]
- Possibly AI-written scriptsAbout 40 custom scripts“roughly 40 custom scripts whose consistent formatting and comments suggested some may have been written with AI assistance”[1]
- Caveat on AI roleNo direct evidence AI was used in the attacks“no direct evidence AI was used to create the exploits or carry out the attacks”[1]
- AttributionPossible Chinese connection, attackers unidentified“Researchers found evidence pointing to a possible Chinese connection, although they have not identified the attackers or confirmed any government involvement.”[1]
- Servers breached25 servers across four countries“Hackers have reportedly broken into 25 servers across four countries”[1]
Sources
- Hackers breached 25 servers, including U.S. government-linked systems — and AI tools may have helpedtomsguide.com · October 28, 2026
How this record is classified. Severity S (4/5): real harm to one organisation, data exposed, or a containment failure. OECD level: incident, an event in which an AI system led to actual harm. Evidence: Reported, meaning press reporting only.
The facts in At a glance and in Every fact and its source are quoted from the sources listed. The summary, What the AI did and What this means for you are written from those sources and checked against them automatically before publication. Records reflect what has been disclosed, not everything that has happened. Incidents are found through news feeds and the GDELT Project.
