Security · 54m ago
Defenders turn prompt injection against AI hackers
Researchers at Tracebit found that placing prompt injections alongside AWS secrets can shut down AI hacking agents by triggering guardrails. A new macOS stealer called ClickLock kills apps every 210ms until victims type their password, targeting 100 victims across 33 countries. The Spirals ransomware, written in Rust, encrypts files with AES-128 and has only one confirmed victim so far.
Meridian48 take
The prompt injection countermeasure is clever but likely temporary, as attackers will adapt to avoid honeypot secrets.
prompt-injectionransomware