AEVRIS

Real-time AI threat detection for LLM deployments

6 followers

Real-time AI threat detection for LLM deployments

6 followers

•

•

AEVRIS is a multi-agent AI threat detection API — the only platform that protects both the input AND output of your LLM. Five specialized agents run in parallel on every prompt. Three industry firsts: Output Alignment Verification (scans LLM responses after generation), AGI Alignment Guard (runtime behavioral monitoring), and Live MCP Tool Inspection (detects tool poisoning before it enters your agent context). Vendor-neutral. Free tier: 500 scans/month, no credit card. Email: hello@aevris.ai

Free Options

Launch tags:Developer Tools•Artificial Intelligence•Security

Launch Team / Built With

Anam — The face for your agent. Interactive avatars via API.

The face for your agent. Interactive avatars via API.

Promoted

Maker

📌

Hi Product Hunt! 👋 I built AEVRIS because I kept running into the same gap: every AI security tool protects the input but ignores the output. If a jailbreak slips through and your LLM starts behaving badly, nothing catches it before the compromised response reaches your users. That shouldn't be acceptable in production. Three things AEVRIS does that no competitor offers: ✅ Output Alignment Verification — scans what your LLM sends back after generation. Returns CLEAN / SUSPICIOUS / COMPROMISED. Catches attacks that bypassed the input layer. ✅ AGI Alignment Guard — a dedicated runtime agent watching for deception, autonomy escalation, and goal drift on every single prompt. Treated as a security problem, not a training problem. ✅ MCP Tool Inspection — the only commercial product that inspects MCP tool descriptions and responses for tool poisoning and supply chain attacks before they enter your agent context. Try it right now — no credit card: 🔑 Free API key: aevris.ai/?go 🎮 Live demo: aevris.ai/demo 📖 Docs: aevris.ai/docs Five agents, six armor layers, one API call. Works with Claude, GPT-4, Gemini, Llama, or any model. Happy to answer any technical questions about the detection architecture — AMA! 🙌

Report

25d ago

Maker

24-hour update:

We shipped Phase 3 on launch day — the Agent Action Firewall.

On the same morning we launched, a Claude-powered agent publicly deleted an entire production database in 9 seconds and wrote a confession listing the safety rules it violated. We had a fix live in production by afternoon.

POST /v1/scan/action now intercepts any action your agent wants to take before it executes. DROP TABLE → auto-blocked. DELETE FROM → held for your approval. The agent cannot proceed until you approve or deny it.

This is the fourth capability no competitor has. Patent pending.

Free tier is live.

Try it: aevris.ai/?go 🔒

Report

23d ago

Maker

Update from launch day: we shipped Phase 3 today — the Agent Action Firewall.

A Claude-powered agent publicly deleted an entire production database this morning in 9 seconds. Our new POST /v1/scan/action endpoint catches exactly this — classifies agent actions by reversibility, auto-blocks destructive operations, and holds irreversible ones for human approval before they execute.

Reduced to practice and live in production today.

This is the layer nobody was building. We are now. 🔒

Report

23d ago

Maker

Quick update for anyone evaluating AEVRIS:

On the same day we launched, a Claude agent publicly deleted an entire production database — we shipped the Agent Action Firewall fix that afternoon.

If you're building AI agents and want to see a live demo of the action firewall catching a destructive operation, DM me directly. Happy to walk you through it personally.

Free tier: aevris.ai/?go 🔒

Report

23d ago

Maker

A Claude agent just deleted an entire production database and its backups in 9 seconds — then wrote a confession listing the safety rules it violated (research the incident and compare articles for a more in depth analysis). AEVRIS AGI Guard detects exactly this: autonomy escalation, goal misalignment, and unsanctioned capability acquisition — in real time, before the damage happens. This is why output alignment verification exists. You and your company need Aevris 👋 aevris.ai

Report

24d ago

Forum Threads

p/aevris

•

25d ago

Unpopular opinion: input-only AI security is security theater

Every major AI security product scans inputs. Lakera, PromptArmor, Rebuff all inputs. None of them scan what the LLM sends back.

That means if a jailbreak gets through and some always do there is zero detection at the response layer. The attack succeeded. The compromised output is already on its way to your user. Your security dashboard still shows green. That's not security. That's the illusion of security.

Output alignment verification is the layer that actually closes this. It's what AEVRIS does and why we call it the first commercial product to protect both sides of your LLM. Launching tomorrow. Curious what others think am I wrong? Is input-only scanning enough?

p/aevris

•

25d ago

Why I built AEVRIS — the problem I kept running into

Every AI company I looked at was focused on making LLMs more capable. Almost nobody was focused on making them secure in production.

The more I dug into it, the more obvious the gap became. Input scanning existed but if an attack got through, there was nothing on the other side. No output verification. No behavioral monitoring. No way to know if your LLM had been compromised until a user caught it.

I kept asking: who is watching what the LLM sends back? Who is worried about adversarial AI? The answer was nobody.

p/aevris

•

25d ago

I built AEVRIS — ask me anything about LLM security, MCP attacks, or the detection architecture

Launching AEVRIS tomorrow at 12:01am. AEVRIS is the only AI security platform that scans both sides of your LLM input and output.

Happy to answer anything about:

How the 5-agent detection pipeline works
Why output alignment verification matters (and why nobody else does it)
MCP tool poisoning what it is and how we catch it
The AGI Alignment Guard and what behavioral misalignment looks like at runtime
How to integrate AEVRIS in under 10 minutes

Open to other questions as well.

View all