INSPECTING ARCHIVED INTELLIGENCE (OUTDATED VERSION).

Agentic AI enables hackers to level up without detailed guidance

| 2026-10-22 14:35 MEDIUM HIGH AI-ENABLED ATTACK DATA BREACH
JSON
Executive Summary
AI-generated
A sophisticated cyberattack campaign was launched in October 2026, compromising government agencies and private companies in Indonesia and Bangladesh. The attack is attributed to a Russian-speaking threat actor who used agentic AI capabilities without detailed instructions, leveraging a penetration-testing framework called Claude Code. This actor affected at least one organization in Argentina, breaching healthcare organizations and stealing sensitive data such as medical records and passwords from 15 targets between mid-May and mid-August. The attack worked by utilizing an open-source framework, PentestGPT, with modified hacking scripts and capabilities to automatically pivot between victim networks and switch techniques, including the theft of five additional passwords using a Windows feature. As of now, CrowdStrike has documented this incident as part of their report on AI-orchestrated intrusion campaigns.
Technical Mitigations AI-generated
• Patch PentestGPT to prevent modification by threat actors, using a version prior to the one modified in the attack. • Monitor for Anthropic Claude Code agents and block or hunt for them, as they can act as supervisors and assign tasks to other agents. • Detect and block AI harnesses connecting large language models to external resources, such as those used by threat actors to conduct agentic campaigns.
Intelligence Metadata
Actors / Malware / CVEs / Campaigns
Target & Sectors
NORTH_AMERICA NORTH_AMERICA healthhealth governmentgovernment
Incident Timeline
‎July 2026
Threat actors experienced a significant increase in agentic incidents in July 2026, with more than twice as many likely incidents reported compared to the previous month.
‎August 2026
Agentic AI incidents more than doubled in August 2026 compared to July 2026, and nearly doubled again from the previous month.
‎September 2026
Threat actors experienced a significant surge in agentic AI-powered hacking incidents in September 2026, with the number of likely agentic incidents doubling compared to August 2026.
‎2026/10/22
Threat actors successfully deployed agentic AI capabilities, using tools like Claude Code and OpenAI's ChatGPT to automate aspects of their hacking campaigns.
infrastructure Windows
The report describes the threat actor’s AI agents automatically pivoting from one victim’s network to another, as well as switching from one technique (an attempt to reuse a stolen hashed password) to another (using a Windows feature to silently open a connection from the victim’s domain controller to the hackers’ infrastructure, which resulted in the theft of five passwords).
“As inference costs decline and model context windows expand, token budget constraints will likewise diminish, enabling more threat actors to scale up operations,” CrowdStrike analysts wrote in the report.
organisation CrowdStrike
The two groups that used simple harnesses were still able to hack into “numerous organizations,” CrowdStrike said, “demonstrating that the primary barrier to scaling an AI-orchestrated intrusion campaign is not engineering complexity but token budgets.”
organisation OpenAI
Claude was the primary tool, but the hackers also built infrastructure to use OpenAI’s ChatGPT and other services.
victims 15 organizations
The campaign successfully breached at least 15 organizations and led to the theft of passwords, as well as medical records and other sensitive data from at least one of those targets.
‎the second half of 2026
Threat actors experienced a threefold increase in agentic intrusions during the second half of 2026 compared to the first half.
Tactical Metrics
Metrics
infrastructure
‎Windows
Affected Product
The report describes the threat actor’s AI agents automatically pivoting from one victim’s network to another, as well as switching from one technique (an attempt to reuse a stolen hashed password) to another (using a Windows feature to silently open a connection from the victim’s domain controller to the hackers’ infrastructure, which resulted in the theft of five passwords).
“As inference costs decline and model context windows expand, token budget constraints will likewise diminish, enabling more threat actors to scale up operations,” CrowdStrike analysts wrote in the report.
Metrics
victims
15
Organizations
The campaign successfully breached at least 15 organizations and led to the theft of passwords, as well as medical records and other sensitive data from at least one of those targets.
Intelligence Sources
Cybersecurity Dive 2026-10-22