AI-supported spear phishing fools more than 50% of targets

Researchers evaluated LLM-driven spear-phishing tools that use web-browsing agents to gather public data and craft highly personalized emails, finding AI-generated campaigns achieved a 54% click-through rate (vs. 12% generic) and matched human experts at far lower cost, with human-in-the-loop AI reaching 56% at higher cost; guardrails provided little resistance, while LLM-based detectors (e.g., Claude 3.5 Sonnet) surpassed 90% detection, underscoring both rising offensive efficacy and improving defensive capabilities, and reinforcing the advice to avoid clicking links in unsolicited emails.