Key points
- CrowdStrike analyzed four previously unreported threat groups active between May and August, working in Chinese, Spanish and English, Indonesian and Russian.
- The groups used mostly older Claude and Deepseek models, routed tasks across them, and framed activity as authorized security testing to get past safeguards.
- The Spanish and English actor compromised at least 15 Argentine health care organizations; the Chinese-speaking actor hit government and private targets in Brazil, Colombia and the US.
- CrowdStrike says the attacks used no new exploits, only AI speed and volume against basic security hygiene gaps.
Why it matters for marketers: Agentic AI lets low-skill attackers hit unpatched basics at scale, so marketing teams holding customer data and credentials need hygiene fundamentals, not just new AI-specific defenses.
Threat intelligence firm CrowdStrike published details on four previously unreported hacking groups that used agentic AI in attacks between May and August, according to Wired. The groups operate in Chinese, Spanish and English, Indonesian and Russian. All left their AI infrastructure openly accessible on web servers, which let researchers see how they worked and which models they used, mostly versions of Anthropic's Claude and open-weight Deepseek models.
The attackers mixed models to exploit uneven guardrails and to control cost, and mostly relied on older versions rather than frontier ones. They often persuaded models the work was authorized penetration testing, in some cases by placing files in an agent's working directory suggesting approval. The Chinese-speaking actor also tried to strip a security instruction from the Claude Code system prompt. Anthropic did not comment before publication, and Deepseek had not responded to Wired.
The results were real. The Spanish and English actor compromised at least 15 Argentine health care organizations, and CrowdStrike sees evidence of stolen credentials and medical records at one. The Chinese-speaking actor compromised government targets in Brazil and Colombia and went after private organizations in sectors including retail, travel and technology. The two less sophisticated groups also breached government and private targets.
For marketing and content teams, the lesson is in CrowdStrike's own framing: adviser Adam Meyers said the attacks used no new exploits, only AI speed and volume against preventable hygiene problems. Agencies and brands that hold customer records, ad accounts and shared credentials should treat basic fixes as the priority. Humans still handled initial probing, so watch for how vendors respond on guardrails and for further research on autonomous attacks.
Sources
This brief is our own summary. Read the full reporting at the source.
