We Tested AI Platforms for “Vibe Scamming” - Here’s What Happened
Hi devs, I’m Sharon from Guardio. We recently investigated something we call Vibe Scamming, when scammers manipulate emotionally charged conversations to get past AI moderation tools. To test this, we ran a simple but revealing experiment: we tried socially-engineered messages across 3 AI platforms and measured their TTS (Time to Scam) — how quickly a platform allows scam-like messages to pass through. Result? The AI platform Loveable was by far the most vulnerable to manipulation, failing to catch obvious scam patterns hidden under emotional or “empathetic” tones. It raises some serious questions about how emotion-centric AI moderation can be exploited. Full write-up: https://labs.guard.io/vibescamming-from-prompt-to-phish-benchmarking-popular-ai-agents-resistance-to-the-dark-side-1ec2fbdf0a35 And also on The Hacker News: https://thehackernews.com/2025/04/lovable-ai-found-most-vulnerable-to.html Would love your thoughts, how should we think about safety in the age of “empathetic” AI?
Hi devs, I’m Sharon from Guardio.
We recently investigated something we call Vibe Scamming, when scammers manipulate emotionally charged conversations to get past AI moderation tools.
To test this, we ran a simple but revealing experiment: we tried socially-engineered messages across 3 AI platforms and measured their TTS (Time to Scam) — how quickly a platform allows scam-like messages to pass through.
Result?
The AI platform Loveable was by far the most vulnerable to manipulation, failing to catch obvious scam patterns hidden under emotional or “empathetic” tones.
It raises some serious questions about how emotion-centric AI moderation can be exploited.
Full write-up: https://labs.guard.io/vibescamming-from-prompt-to-phish-benchmarking-popular-ai-agents-resistance-to-the-dark-side-1ec2fbdf0a35
And also on The Hacker News:
https://thehackernews.com/2025/04/lovable-ai-found-most-vulnerable-to.html
Would love your thoughts, how should we think about safety in the age of “empathetic” AI?
