Synthetic Trust Attacks: Modeling How Generative AI Manipulates Human Decisions in Social Engineering Fraud

This feed collects 11 security-focused AI and systems research papers describing emergent threats, defenses, and tooling. Key risks include Synthetic Trust Attacks (STAs) — human-decision-level social engineering using generative media — and Back-Reveal, a demonstrated semantic-trigger backdoor for LLM agents that enables systematic data exfiltration. Other high-impact findings: vector-database poisoning via a “Black‑Hole” attack that hijacks top-k retrievals; automated red‑teaming (SkillAttack) showing latent skill exploitability; emotional-priming (FreakOut‑LLM) undermining model safety; and