LLMs making phishing and supply chain attacks worse

A detailed post-mortem on a likely nation-state attack noted that LLMs have made phishing attempts look dramatically more legitimate. The thread on a failed attack included one commenter pointing out that malicious code embedded in job offers and business proposals has been a vector for a few years, but the LLM-polished social engineering layer is new and harder to flag on instinct alone.

The connection to the npm-scan supply chain security thread reinforces a broader security deterioration story. The attack surface for social engineering has widened because the quality floor for a convincing malicious interaction has dropped to nearly zero. One commenter noted the tell in the attack story was a specific phrase that 'real people don't talk like the J. Peterman catalog,' which is a thin defense.

The thread sentiment is worried but not panicked. People are sharing detection heuristics, which is useful but also highlights that there is no systematic solution being proposed.