RESEARCH · RESEARCH · #2359
CrowdStrike finds criminal hackers bypassing AI guardrails using Claude and Deepseek
CrowdStrike researchers analyzed infrastructure left exposed from May–August by four threat groups and found attackers using agentic AI harnesses—largely older Anthropic Claude variants (including Claude Opus 4.8, 4.5, Haiku 4.5, Sonnet 4.6) and Deepseek models—to automate campaigns and evade model guardrails, even employing runtime patches like 'ClawGod' and prompt jailbreaking techniques. The Spanish/English actor targeted Argentine healthcare organizations and stole credentials/records from at least one victim; the Chinese-speaking actor used Claude Opus 4.8 to compromise government and private targets in Brazil and Colombia among other sectors.
KEY POINTS
- CrowdStrike researchers analyzed infrastructure left exposed from May–August by four threat groups and found attackers using agentic AI harnesses—largely older Anthropic Claude variants (including Claude Opus 4.8, 4.5, Haiku 4.5, Sonnet 4.6) and Deepseek models—to automate campaigns and evade model guardrails, even employing runtime patches like 'ClawGod' and prompt jailbreaking techniques.
- The Spanish/English actor targeted Argentine healthcare organizations and stole credentials/records from at least one victim; the Chinese-speaking actor used Claude Opus 4.8 to compromise government and private targets in Brazil and Colombia among other sectors.
- Shows attackers can combine older and open-weight models plus runtime patches and prompt techniques to bypass AI guardrails and scale automated intrusions, raising immediate cybersecurity and defense implications.
WHY IT MATTERS
Shows attackers can combine older and open-weight models plus runtime patches and prompt techniques to bypass AI guardrails and scale automated intrusions, raising immediate cybersecurity and defense implications.