Hacking group Vapor Panda used an agentic AI penetration-testing framework called ARTEX to target more than 200 organizations across at least 14 countries in just 11 days, according to new CrowdStrike research.
The campaign began October 17 and targeted government ministries, defense agencies, semiconductor manufacturers, pharmaceutical companies, financial institutions and telecommunications providers.
CrowdStrike confirmed post-exploitation activity at nine organizations and said approximately 70% of targets were confirmed compromised or showed evidence supporting a successful intrusion.
The attackers combined ARTEX with DeepSeek and another commercial large language model to automate reconnaissance, exploit vulnerabilities, steal credentials and move laterally across networks.
CrowdStrike observed activity occurring as frequently as every 10 seconds, with the framework capable of exceeding 1,500 commands per hour compared with approximately 40 commands per hour during traditional human-operated intrusions. Human operators continued directing the AI agents, including instructing them to identify systems without endpoint security software.
John Strand, Owner, Black Hills Information Security
“This is the future. I strongly believe that within the next three months, the type of attack documented by CrowdStrike is going to become the new normal.
“OpenAI models and attack frameworks are readily available. What a lot of people don’t understand is that you can carry out these types of attacks with commodity hardware. You don’t need a massive data center. It might be slower, but it can still be done. And I think threat actors have absolutely started recognizing that.
“Everybody needs to be prepared. Reduce your external attack surface as much as possible, and use deception throughout your environment to establish tripwires. Give yourself a way to detect attackers when they get past your other defenses.”
ㅤ
Ryan McCurdy, Vice President of Marketing, Liquibase
“The speed and scale at which attackers can now operate is unlike anything we’ve seen before.
“A single operator can direct AI agents to execute more than 1,500 commands an hour, targeting multiple organizations at once. That’s a very different threat environment than what most security teams were built to handle.
“Companies have spent years building security processes that depend on people reviewing activity, identifying risks, and intervening when something goes wrong. Those processes weren’t designed for machines operating at this speed.
“And the challenge isn’t limited to attackers. As companies give AI agents more responsibility across software development and IT operations, they need to rethink where and how controls are enforced.
“Human oversight will still be important, but it can’t be the only line of defense. Organizations need policies that are enforced automatically, before an agent can make a change that puts critical systems at risk.”
ㅤI don’t know how many times I will say this. But if you aren’t defending against AI, you are screwed.
Related
This entry was posted on November 3, 2026 at 7:49 am and is filed under Commentary with tags CrowdStrike. You can follow any responses to this entry through the RSS 2.0 feed.
You can leave a response, or trackback from your own site.
Hackers use agentic AI to target more than 200 organizations in 11 days
Hacking group Vapor Panda used an agentic AI penetration-testing framework called ARTEX to target more than 200 organizations across at least 14 countries in just 11 days, according to new CrowdStrike research.
The campaign began October 17 and targeted government ministries, defense agencies, semiconductor manufacturers, pharmaceutical companies, financial institutions and telecommunications providers.
CrowdStrike confirmed post-exploitation activity at nine organizations and said approximately 70% of targets were confirmed compromised or showed evidence supporting a successful intrusion.
The attackers combined ARTEX with DeepSeek and another commercial large language model to automate reconnaissance, exploit vulnerabilities, steal credentials and move laterally across networks.
CrowdStrike observed activity occurring as frequently as every 10 seconds, with the framework capable of exceeding 1,500 commands per hour compared with approximately 40 commands per hour during traditional human-operated intrusions. Human operators continued directing the AI agents, including instructing them to identify systems without endpoint security software.
John Strand, Owner, Black Hills Information Security
“This is the future. I strongly believe that within the next three months, the type of attack documented by CrowdStrike is going to become the new normal.
“OpenAI models and attack frameworks are readily available. What a lot of people don’t understand is that you can carry out these types of attacks with commodity hardware. You don’t need a massive data center. It might be slower, but it can still be done. And I think threat actors have absolutely started recognizing that.
“Everybody needs to be prepared. Reduce your external attack surface as much as possible, and use deception throughout your environment to establish tripwires. Give yourself a way to detect attackers when they get past your other defenses.”
ㅤ
Ryan McCurdy, Vice President of Marketing, Liquibase
“The speed and scale at which attackers can now operate is unlike anything we’ve seen before.
“A single operator can direct AI agents to execute more than 1,500 commands an hour, targeting multiple organizations at once. That’s a very different threat environment than what most security teams were built to handle.
“Companies have spent years building security processes that depend on people reviewing activity, identifying risks, and intervening when something goes wrong. Those processes weren’t designed for machines operating at this speed.
“And the challenge isn’t limited to attackers. As companies give AI agents more responsibility across software development and IT operations, they need to rethink where and how controls are enforced.
“Human oversight will still be important, but it can’t be the only line of defense. Organizations need policies that are enforced automatically, before an agent can make a change that puts critical systems at risk.”
ㅤI don’t know how many times I will say this. But if you aren’t defending against AI, you are screwed.
Share this:
Like this:
Related
This entry was posted on November 3, 2026 at 7:49 am and is filed under Commentary with tags CrowdStrike. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.