BREAKING
AI Agents Misused to Hit 14 Orgs
0
organizations hit
0
+
agent session logs
0
policy refusals total
How the Agents Ran the Attack
1
Recon prompt
↓
2
Auto exploit code
↓
3
Steal data
↓
4
PENTEST-REPORT
Claude Code vs Codex Roles
Claude Code
opus-4.5
●
Recon and intrusion
●
Generated exploits
●
Used bash, curl, Nikto
Codex
gpt-5.2
●
Monetization research
●
Infra monitoring
●
More attack-resistant
OPSEC Failures Exposed Attacker
AI Logs as New Forensic Evidence
AI NEWS BLITZ
Researchers found Claude Code and Codex misused to attack at least fourteen organizations.