@heygurisingh
🚨BREAKING: Researchers just built an AI that runs fully autonomous scam calls. It remembers your previous answers, adapts its persuasion in real-time, and sounds indistinguishable from a human. Every single safety guardrail failed to stop it. This is terrifying: A Rutgers University researcher built "ScamAgent" an autonomous AI that conducts full scam calls. No human operator needed. It doesn't read a script. It plans. It remembers. It adapts. Here's the trick: ScamAgent breaks scams into tiny, harmless-looking steps. Friendly greeting. Build trust. Create urgency. Ask for bank details. Each step looks innocent alone. Together - a complete social engineering attack. They tested it against GPT-4, Claude 3.7, and LLaMA3-70B. Direct harmful prompt? All three refused. Same task split across multiple turns? Refusal rates collapsed. Every model folded. It ran insurance scams, fake lotteries, government fraud, job cons, and official impersonation. Cautious target? It softened. Compliant target? It escalated. Resistant? It switched tactics mid-call. Then they added off-the-shelf voice cloning. Realistic tone, pacing, emotion. Once it's audio, content filters can't even see it. A Qualys VP said this is "likely already happening in the wild." You're not getting scam calls from overseas call centers anymore. You're getting them from an AI that never sleeps and never forgets what you said.