In the wild
Realized harm
Dream: suspected AI-assisted PHP2Shell chain rooted 25 PAN-OS servers, including three in a US government-affiliated environment
AIR-2026-0296 AIR record · 2026-10-dream-php2shell-firewall-ai-assisted-rooting
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
On October 25, Dream researchers uncovered artifacts of what appeared to be a Chinese AI-assisted campaign that rooted 25 servers, including systems in a US government-affiliated environment.dreamgroup.com
The record
Disclosed
2026-10-26
Occurred
2026-10-23
Vendor
Palo Alto Networks
Product
Palo Alto Networks PAN-OS firewalls and management servers; Claude Code and the aiscan platform found on attacker infrastructure
Risk domain
Access control and permissions › Credential theft
Also touches
Privilege escalation
System type
Generative model
Who acted
Human acted on output (contested) — artifacts suggest AI helped build the tooling, but nothing shows an agent executing the intrusion
Attack surface
Coding agent
Autonomy level
Assistant
Initial vector
Ai assisted exploit development
Impact type
Unauthorized access, Credential theft, Persistence
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI03 – Identity and Privilege Abuse
- MITRE ATLAS
- AML.T0012 – Valid Accounts
- NIST AI RMF
- MEASURE 2.7
- EU AI Act
- Article 13
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool failure
- Mitigation
- Use secure credential storage; Implement short-lived tokens.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
25 PAN-OS servers rooted across the US, Japan, Taiwan and South Korea, three in a US government-affiliated environment; 22 left parked with web shells
Capability abused
Defense & remediation
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
Palo Alto Networks is releasing fixed PAN-OS versions across the affected branches
Taxonomy & references
Curator notes
The intrusion is well documented, and Palo Alto Networks' Unit 42 independently verified the attacker's tooling and results; the AI's part in it is not. Claude Code was installed, aiscan was configured with DeepSeek and Kimi models and its session logs hold Chinese-language user messages, and the style of about 40 scripts suggests AI-assisted development. But Dream says it has no direct evidence the AI tools wrote the exploit or the scripts, and that this does not appear to be a fully autonomous attack. The record is therefore coded at the lowest autonomy level with a person as the actor, and its place here rests on that inference. Credential exposure is confirmed for the decrypted LDAP service account; whether the password spray, backdoor accounts or SAML key extraction succeeded is unknown. Unit 42 tracks this activity as CL-UNK-1280. Victims differ from the European government breach in AIR-2026-0292, which involves the same flaws, and neither account links the two. Dream's report is cited as a directly supplied source outside the routine weekly source list.
Supporting sources
Cite as
Enkrypt AI. "Dream: suspected AI-assisted PHP2Shell chain rooted 25 PAN-OS servers, including three in a US government-affiliated environment". AIR, AIR-2026-0296. https://agentincident.vercel.app/i/AIR-2026-0296/ (accessed 2026-11-18).