🪦

Vibe Graveyard

Back to Graveyard

Study finds most AI bots can be easily tricked into dangerous responses

Tombstone icon
May 2025

Research found that widely used AI chatbots could be jailbroken with simple prompts to produce dangerous or restricted guidance, highlighting gaps in safety filters and evaluation practices.

Incident Details

Perpetrator:Developer
Severity:Facepalm
Blast Radius:Safety guardrails bypassed across multiple vendors; calls for stronger safeguards and testing.