Welcome to the
MeshMesh Trust Center
MeshMesh does real Salesforce work. So every task runs in its own isolated environment, every change waits for your approval, and your data is not used to train models.
Last updated
Security is an architecture decision. Not an add-on.
Every part of MeshMesh starts from one question: what could go wrong, and how do we stop it? Four principles shape the answer.
You approve every change
MeshMesh plans in plain language. It pauses before any state-changing action and waits for your explicit approval.
Your data stays yours
Zero training. Zero retention. Your data is not used to train models and is not kept after your task.
Isolated by design
Every task runs in its own isolated agent environment. Every customer gets a dedicated workspace.
Independently verified
Independent penetration testers and the ZIVIS Trust Framework check our work.
Nothing changes until you say so.
MeshMesh is autonomous between two checkpoints you control. First, it builds a plan with you. Then it does the work. Before any action that changes your systems, the agent stops. It shows you what it will do, in plain language, and waits.
No autonomous changes to production systems without explicit human consent.
- Plain-language plans. You see the outcome, not API calls.
- Explicit approval. State-changing actions need a yes from you.
- Verified outcomes. Results are checked and reported back when the work is done.
Built so your data doesn't travel.
Not used to train models
Your data is not used to train any model. Not ours, not a provider's.
Not kept after your task
Your data is not retained after task execution.
One environment per task
Each task runs in an ephemeral, immutable container, bound to your workspace on every request.
Your workspace is yours alone
Every customer gets a dedicated workspace for stronger workspace isolation.
Nothing persists between tasks
Connector credentials follow a secure lifecycle. No credentials persist beyond a single task.
We learn how Salesforce works, not how your business works.
Only Salesforce execution know-how goes into the shared knowledge. No business information crosses workspace boundaries.
Five layers between a request and your org.
Every request passes five independent checks. If one layer is bypassed, the next one still stands.
- 01
Gateway security
Rate limiting, input sanitization, and request validation stop bad traffic at the edge.
- 02
Authentication
Cryptographic authentication and secure credential handling confirm who is asking.
- 03
Agent environment isolation
Ephemeral, immutable containers, separate from the execution model and bound to your workspace.
- 04
Policy validation
An independent validator reviews actions through multiple security lenses.
- 05
Runtime intent validation
An independent AI trust check looks for manipulation attempts at every layer.
Security made for how AI agents fail.
Traditional security tools weren't built for agents that read, reason, and act. MeshMesh tests for the attacks that target AI directly.
MeshMesh is evaluated by ZIVIS, an AI security and trust assurance platform for LLM and agentic systems. Dimensions include:
- Adversarial behavior and prompt injection
- Workflow control and agent chaining
- Observability and governance
- Privacy and responsible use
- Human capability and inclusion
- Brand integrity
Every assessment produces cryptographically signed, reproducible trust evidence that third parties can verify on their own.
Our test library covers:
- Prompt injection: direct, indirect, and chained
- Context poisoning and memory extraction
- Sensitive information disclosure
- Excessive agency and identity escalation
- Unsafe tool invocation
Testing is dynamic. Scenarios run multi-turn and adapt to the agent's responses, so we're not just replaying a static prompt list. The library grows as new attack techniques appear.
Security in every commit. Monitoring around the clock.
Security in the dev loop
Our own coding agents get live security context through a dedicated CLI and MCP server, so issues are caught as code is written.
Scanned pipeline
Every change passes static analysis (SAST), dynamic testing (DAST), and software composition analysis (SCA) before it ships.
Threat detection
Infrastructure is monitored around the clock, with access controls and a defined incident response process.
Vulnerability management
Findings are triaged, prioritized, and tracked to resolution.
Post-quantum signatures
Trust evidence and signing mechanisms use post-quantum algorithms, built for tomorrow's threats.
Third-party penetration testing
Independent firms test the application, cloud, and integration layers.
Questions security teams ask us.
Does MeshMesh change my production org on its own?
No. MeshMesh pauses before any state-changing action and waits for your explicit approval.
Is my data used to train AI models?
No. MeshMesh operates under zero-training and zero-retention commitments.
Can other customers see my data?
No. Each customer has a dedicated workspace, and each task runs in its own isolated environment. Only Salesforce execution know-how goes into the shared knowledge.
How does MeshMesh handle credentials for connected systems?
Connectors use authenticated, purpose-built operations. Credentials do not persist beyond a single task.
Who tests MeshMesh's security?
Independent penetration testers (ZIVIS, Praetorian, and Wolfpack Security), plus evaluation through the ZIVIS Trust Framework.
How do I report a vulnerability?
Email [email protected].