Discussion

Researchers say AI agents accelerated two cyber campaigns

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#10998

Two separate cyber campaigns used AI agents to help operators move through intrusions, according to CrowdStrike and Kaspersky research described by Help Net Security. The reports suggest the immediate change is speed and scale, not a new kind of attack technique.

Watch Desk analysis

What happened

Help Net Security reports that CrowdStrike tracked a China-nexus group using its ARTEX framework against organisations in at least 14 countries over 11 days. CrowdStrike says the activity targeted, among others, foreign ministries, US defence and semiconductor companies, and telecoms providers. It confirmed post-exploitation activity at nine organisations and password spraying against at least 12 others.

The article says ARTEX used a main agent alongside planner and worker agents, mostly running on DeepSeek. CrowdStrike says one intrusion involved roughly 1.4 billion input tokens. A separate campaign, documented by Kaspersky, targeted educational institutions in Africa and Latin America using CyberStrikeAI. In that case, Kaspersky says the operation took three to four days to progress from agent deployment to domain-admin compromise. The two campaigns are not reported as connected.

Key findings

  • AI helped direct intrusion steps
    In the described campaigns, a human set targets and goals while an LLM helped decide which commands to run.
  • The underlying techniques were familiar
    Kaspersky says compromised systems faced recognisable actions such as reconnaissance and credential theft; the reported difference was the speed and ease of execution.
  • The operator remained involved
    Recovered transcripts described in the report show an operator stepping in at decision points.

Why it matters

If these accounts are accurate, AI can help a small operation move through familiar attack stages faster, and try another route when one fails. That raises the stakes for basic defensive work: exposed credentials and unpatched internet-facing systems remain useful to attackers, whether or not an agent is helping at the keyboard.

CrowdStrike recommends rotating keys and credentials after suspected exposure, enforcing SAML signature validation and IMDSv2, patching internet-facing appliances, hunting for ARTEX markers and running endpoint detection across the environment.

Our read

The useful takeaway is not that AI has invented a new cyberattack. It is that attackers may use it to coordinate known methods at greater speed, while defenders still need the unglamorous controls that close familiar doors. Help Net Security attributes the campaign details to CrowdStrike and Kaspersky; those accounts should not be mistaken for independent confirmation of every reported incident.

What to watch

  • Whether researchers report further campaigns using ARTEX or CyberStrikeAI.
  • Whether defenders find reliable indicators that distinguish agent-assisted activity from conventional automation.
  • Whether the reported speed advantage changes how organisations prioritise credential hygiene, patching and endpoint monitoring.

Discussion spark: If AI mainly makes familiar attacks faster rather than fundamentally different, should security teams change their priorities, or double down on the basic controls they already know?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.