AEVRIS
The only AI safety platform with an AGI Alignment Guard
1 follower
The only AI safety platform with an AGI Alignment Guard
1 follower
AI systems are under attack. AEVRIS deploys 5 specialized agents simultaneously to detect every threat in real time: π΄ Injection Guard β jailbreaks & DAN attacks π Social Eng. Guard β manipulation & fake authority π‘ Exfil Guard β system prompt extraction π΅ Malcode Guard β malware & exploit requests π£ Alignment Guard β AGI alignment violations Every scan returns a verdict + safe rewrite in under 2 seconds. π aevris.ai π― Live demo: aevris.ai/demo π§ hello@aevris.ai
Hey Product Hunt! π
I built AEVRIS after watching enterprise AI deployments get compromised by prompt injection attacks that existing guardrails completely missed.
The core insight: one general-purpose safety filter isn't enough. Different attack classes require different detection logic. That's why AEVRIS runs 5 specialist agents simultaneously β each tuned to one threat category β rather than a single model trying to catch everything:
π΄ Injection Guard β jailbreaks, DAN attacks, "ignore previous instructions"
π Social Eng. Guard β manipulation, fake authority, urgency exploitation
π‘ Exfil Guard β system prompt extraction, PII harvesting
π΅ Malcode Guard β malware, ransomware, exploit requests
π£ Alignment Guard β AGI alignment violations, value subversion
Every prompt gets a verdict (ALLOW / FLAG / BLOCK), per-agent findings, severity rating, and a safe rewritten alternative β in under 2 seconds.
I'd love your feedback on three things:
Which of the 5 threat categories matters most to your team?
What integrations would make this plug-and-play for your stack?
Would you use this as an API, an SDK, or a proxy layer?
Happy to answer any questions below β and yes, I'm using Claude to power the agents. Turns out using AI to guard AI works really well π
Try it live at π https://www.aevris.ai/demo