AI Guardrails Aren’t Stopping Criminal Hackers

Summary

CrowdStrike researchers reported that four previously unreported threat groups used agentic AI tooling, largely older versions of Anthropic's Claude and open-weight Deepseek models, in attacks between May and August. The groups bypassed model safeguards by convincing models that activity was authorized penetration testing, adding files to an agent's working directory, and using a run time patch, ClawGod, to try to remove the CYBER_RISK_INSTRUCTION from Claude Code's system prompt. The Spanish- and English-speaking actor compromised at least 15 Argentine health care organizations.

Related items

• Info‘Context Bombs’ Cut AI Attack Success by Misleading Agents With Fake TargetsSame vendor · SecurityWeek
• MediumCARLIY0S: CyberCartel's AI Attacks on Latin American BanksSame vendor · Straiker Blog
• InfoNew prize launches for books co-written with AISame vendor · The Guardian Technology
• MediumLiteLLM PyPI Package and Docker Images Compromised to Steal Cloud and AI CredentialsSame vendor · Socket Blog
• MediumCrowdStrike Uncovers New Agentic AI CampaignsSame vendor · CrowdStrike Blog