provnai

Verifiable trust layer for autonomous agents | VEX Protocol 🦀 | McpVanguard🛡️Real-time guardrails and cryptographic accountability

[provnai @provnai]

MCP security is not just prompt safety.

It is an execution-boundary problem.

The dangerous moment is often not what the model says.

It is what the agent is about to do.

[provnai @provnai]

Agents should not turn model intent into privileged action without runtime enforcement.

Research note:
https://t.co/WIb3bAtF6H

Repo:
https://t.co/UBVdhWo0WC

Looking for critique from MCP builders, agent framework teams, and AI security engineers.

[provnai @provnai]

MCP security is not just prompt safety.

It is an execution-boundary problem.

The dangerous moment is often not what the model says.

It is what the agent is about to do.

[provnai @provnai]

McpVanguard ships monitor, balanced, and strict profiles because runtime security has tradeoffs.

Stricter enforcement can block legitimate admin, research, and incident-response workflows.

That tradeoff should be visible.

[provnai @provnai]

Signing proves origin. Not safety.

McpVanguard 2.0.0-rc1 is live — a 3-layer security firewall for MCP that catches prompt injection, exfiltration, and behavioral attacks before they reach your tools.

Zero config. Claude Desktop + Cursor ready

https://t.co/UBVdhWo0WC

#MCP #AI

[provnai @provnai]

The crisis of the future is not intelligence.

It is self-authorization.

🛡️

[provnai @provnai]

Good point. CHORA-VEX is much stronger on the execution side than on the perception side.

If an agent is shown poisoned or selectively manipulated input, that can still distort planning or intent formation. What our boundary changes is that structurally plausible reasoning still does not automatically become executable action.

[provnai @provnai]

@AnthropicAI If an internal state is no longer admissible, execution shouldn’t just be flagged. It should be unreachable.

[provnai @provnai]

@github Governance Layer is missing 🧠

[provnai @provnai]

@SeiNetwork The future is cognitive and deterministic 🧠

[provnai @provnai]

Feels a lot more like the agentic future we’re actually building.
If you’ve been following the madness, go have a look and tell me what you think → https://t.co/8fz2Oq0Prd

#VEXProtocol #AgenticAI #OpenSource #provnai

[provnai @provnai]

Love the framing – 'HELM' as the kernel below is exactly right. 🧱

McpVanguard is just the shield, but we've been building the VEX Protocol to handle that 'signed receipt per call' logic at the enforcement level. We're calling it the AEM (Action Enforcement Module).

Good to see others naming the gap!

[provnai @provnai]

AI agents now have access to your filesystem, shell, databases, and APIs.
The MCP ecosystem just crossed 10,000+ public servers.
There is no enforcement boundary between what an agent intends to do and what actually runs.
This is the problem nobody is talking about. 🧵

[provnai @provnai]

@SeiNetwork @Morpho ⚓️

[quinten @quint3ns]

1-click governed AI agent stack.

VEX handles trust and verification. CHORA handles governance and escalation. Railway makes it deployable in seconds.

Almost there. #CHORA #AI @provnai

[provnai @provnai]

@yq_acc @AnthropicAI The lack of any built-in quality or security scan on the registry side feels like the next big incident waiting to happen.

[provnai @provnai]

@berman66 @1Password One breach and every connected system lights up. Most teams are still thinking static API keys when agents are basically dynamic sessions now.

The playbook they’re sharing at RSA next week sounds like the shift a lot of security teams are about to wake up to.