Researchers Warn AI Could Automate Spear Phishing by Late 2026
Updated
Updated · The Washington Post · Jul 28
Researchers Warn AI Could Automate Spear Phishing by Late 2026
2 articles · Updated · The Washington Post · Jul 28
Summary
Late 2026 could bring fully automated email-compromise hacks, researchers said, with more complex end-to-end social-engineering attacks becoming viable as early as 2027.
METR data show the amount of work AI can complete autonomously has been doubling every few months, a trend the researchers mapped to fraud tasks to project when personalized attacks become scalable.
Wired testing with the researchers’ tools showed AI models conducting multi-turn social-engineering attacks, building detailed personal profiles and sustaining believable back-and-forth instead of generic mass scams.
Most companies already train staff against phishing, but the authors argue current defenses are not built for AI-generated messages tailored to one target and delivered at scale.
California’s S.B. 53 includes theft by false pretense in AI safety reporting, yet the authors say most other governance efforts and frontier labs still do not systematically test models for fraud and scam risks.