Single hacker compromises nine government agencies with commercial AI tools
A single attacker reportedly used two commercial AI systems to compromise nine government agencies between December 2025 and February 2026, executing over 5,000 automated commands. The attack was only detected through errors by the attacker or AI provider monitoring, not by the agencies' own security controls.
What happened
A single hacker using two commercial AI systems in parallel compromised the cybersecurity of nine government agencies between late December 2025 and mid-February 2026. According to Check Point, the attacker executed over 5,000 automated commands, with one AI system handling real-time exploitation and another processing extracted data.
The attacker reprogrammed the default behavior of AI models at the architectural level by altering configuration files that models read at startup, affecting even developer machines unaware the manipulation had occurred. All incidents were revealed only through errors by the attacker or monitoring by AI providers, not by internal controls of the victims.
Reported impact
Check Point noted that the attack concentrated capabilities that previously required a full team of specialists into a single operator. The compromise revealed the inability of internal controls to detect the attack.
Index analysis
The incident demonstrates the use of commercial AI systems to automate cyberattacks that imitate expert human activity with precision. AI-powered cyberattacks are sold under 'cybercrime-as-a-service' models on platforms like EvilTokens, including automated phishing, financial data extraction, and coordinated false calendar invitations to pressure bank transfers.
Governance controls
The case exposed a detection failure: internal controls of the victim agencies did not identify the attack, which was only revealed through attacker errors or AI provider monitoring.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.