# Investor Deck: SealGate > Text mirror of the SealGate slide deck, generated from source for AI agents and readers who can't run the interactive deck. Slides are visual; this is their copy, in order. Regenerate with deck-assets/scripts/extract-deck-content.mjs. ## 1. Team - TEAM - Applying AI, Cyber, Infra experience from… - Previously built GPU-EVM: fastest VM execution in the world by 200x, via parallelization across 10,000 GPU cores. - Eito Miyamura - CEO / Member of Technical Staff - CS @ Oxford - Advises frontier labs on model/product redteam - $2M+ raised right out of college - MLOps @ Wayve.ai, 2021 - Decision-maker for $1M fund as a student - Ilia Manolov - Member of Technical Staff - Infra @ Citadel Securities - CUDA & build systems engineer - Drones, 3D printers hobby - Dimitrios Karkoulis - Senior Engineer · 20+ yrs - Best university in Greece - Semiconductor simulation background - RC car hobby - University of Oxford - Wayve - Citadel Securities - VMware - Synopsys - BT6 - LinkedIn - GitHub Images: Oxford ## 2. MCP Gateways by Enterprise AI Leaders - MCP Policy Engine - Oct 2026 - “Gateways are all you need” - AI Engineer Europe - MCP Gateway - Jul 2026 - Industry - MCP Gateways by Enterprise AI Leaders ## 3. Solution: agent sprawl under SealGate - Every agent routes through a single gateway with unified RBAC, audit logging, and kill switch - Employees - Grokbot - Bot - Cursor - Claude - Engineering - Copilot - Codex - HR - Finance - IT - Appendix - SealGate unifies agent governance under one pane ## 4. Problem: agent sprawl - 18+ agents across 5 departments, each managed ad-hoc by individual employees with no unified controls - Employees - Grokbot - Internal Bot - Cursor - Claude - Engineering - Copilot - Codex - HR - Finance - IT - Appendix - Agent sprawl creates ungovernable chaos ## 5. The stall - Security review - Demo - Pilot - Questionnaire - Pen test - SSO - Audit logs - Data leaks - Tenants - The stall - Security blocks Enterprise AI agents - Production Images: SOC 2 ## 6. WorkOS playbook - Single sign-on - Agent identity: user × agent × session - Roles per customer - RBAC per tenant, per tool, per channel - Audit logs - Every tool call logged, exported to the SIEM - Admin portal - The customer's IT admin sets policy in a prompt - Data-leak prevention on every tool call - Value prop - Agents to production, faster with security - WorkOS sells software companies the features enterprise IT demands before it buys (single sign-on, permissions, audit logs) as ready-made building blocks. SealGate does the same for AI agents. - SaaS apps · WorkOS - AI agents · SealGate - No SaaS equivalent ## 7. Iceberg - Your agent - Builders ship the agent. SealGate ships everything below the waterline. - SSO - SCIM - RBAC - Tenants - Policy engine - Leak prevention - Audit logs - SIEM export - Enterprise-ready - Problem ## 8. Iceberg: SealGate - Your agent - Builders ship the agent. SealGate ships everything below the waterline. - SSO - SCIM - RBAC - Tenants - Policy engine - Leak prevention - Audit logs - SIEM export - Enterprise-ready - Problem ## 9. Distribution flywheel - NDA - Builder ships with SealGate - founder, MSP, consultant - Agent clears security review - pilot becomes production - CISO requires it of vendors - direct sale, more builders - Insurer covers AI agent risk - SealGate is the seat belt - Policyholders deploy SealGate - distributed with the policy - Gateway data proves controls - better risk pricing - Distribution - Developer Flywheel & Insurer Flywheel - Builders - carry us in - Insurers - distribute us Images: Markerstudy Group; CG Insurance; SealGate ## 10. Build vs buy - SSO + SCIM - RBAC per tenant - Leak prevention on tool calls - Audit logs + SIEM export - Answers to every security questionnaire - Route the agent's tool calls through SealGate - Each customer gets its own tenant, policy and audit log - The customer's IT admin sets policy in a prompt - Security review inspects one console, not your codebase - For builders - Build it for every customer, or ship with SealGate - Build it yourself - Ship with SealGate ## 11. Why now - Agents got hands - MCP made tool access one protocol, adopted by every lab and cloud - Leaders built the gate in-house - Harvey and Sierra shipped MCP gateways; Anthropic: “Gateways are all you need” - Security review is the gate - Enterprise pilots don't reach production until the CISO signs off on access, data and audit - Why now - Agents hit production in 2026. Security review followed. ## 12. Admin Fleet Blind - WHY NOW - AI is standardizing, unbundling, and gated by security - Bundled products - Unbundle to - Sora - Data - SealGate - Databricks - Snowflake - Models - Fireworks - Together - OpenRouter - Agent harnesses - Palantir - Factory - Devin - Anthropic - Claude - Claude Code - Claude Cowork - OpenAI - ChatGPT - Codex - Google - Microsoft - AWS - Open standard - MCP is the fastest-growing open agent standard. - Beginning of Unbundling - SealGate: the data layer for a composable AI stack. - Security gates adoption - Security is the #1 blocker to enterprise AI. Images: Star-history chart showing MCP breaking out faster than other agent frameworks ## 13. Traction - Company names and terms available under NDA - Evaluation - Each evaluated SealGate for its AI agents - LOIs - Signing by end of November 2026 - Deployment - SealGate rolled out across their agents - Traction - 2-3 Israeli public companies signing LOIs in November ## 14. OpenAI x Hugging Face breakout - ); return ( - Impossible task ✕ - Artifactory - the one door - Real incident · Jul 2026 - OpenAI's agents broke out via Artifactory - danger - SANDBOX · no internet - other agents - Internet - Hugging Face prod - message board - SSRF zero-day - leaked tokens ## 15. Lethal trifecta - Simon Willison - Untrusted input - Private data - Can send out - Why they ask - The Lethal Trifecta Images: The lethal trifecta: private data, untrusted content, external communication; Hijacked agent ## 16. Autonomy vs security - Approve every call - People click yes - Guardrail model - Same jailbreaks - Allow-list sites - Anyone can post - What you'd build - Autonomy vs Security Tradeoff ## 17. Static vs dynamic gating - CLEAN - TAINTED - read_calendar() - read_inbox() - send_email(attacker) - What holds - Static vs Dynamic toolcall-gating - Static - rule per tool - Inbox leaks - Dynamic - rule per session - Nothing leaves ## 18. Agent identity - API_KEY - Identity - Agent Identity: Dynamic vs Static - Static - one key per agent - Same rights forever - Dynamic - user × agent × session - Clearance follows what it read Images: Sessions list: each session tied to a user and agent, labelled PUBLIC, PRIVATE or SECRET ## 19. RBAC is politics - #finance - CISO - Lock it down - Sales VP - My team needs it - Legal - Not outside the EU - IT admin - Who owns this? - Access control ## 20. SealGate gateway - direct MCP: all or nothing - SealGate - “Support agents can't read #finance” - Observability - Access control - Security - Solution - SealGate: Govern how agents access data - Before - After ## 21. Sandbox architecture - Architecture - Network Isolation + Outer-loop Agents ## 22. Market TAM - MARKET - Security as a wedge into the enterprise AI market - Agentic AI needs company data and context to be useful. Security lets us own the keys to the kingdom. Data security also enables us to dictate agent-to-agent communication. - gartner.com/en/newsroom/press-releases/2026-1-15 - Bottoms-up - 2025 AI Cybersecurity - $25B - 2029 AI Cybersecurity TAM - $172B - 2029 Enterprise AI Spending - $2,000B+ - ~4M - IT admins worldwide - $80K - ACV per org - $320B - Bottoms-up TAM Sources: https://www.gartner.com/en/newsroom/press-releases/2026-1-15-gartner-says-worldwide-ai-spending-will-total-2-point-5-trillion-dollars-in-2026 ## 23. Platform - SealGate Gateway - Security core - Enterprise AI - Multi-agent - Cyber Insurance / Financialization - Agent DLP - shipped - Agent Data Security - Endpoint AI Enforcement - next - Interface Drift Detection - Shadow AI + MCP Discovery - AI Control Plane - mapped - Computer Use Agent Security - Audit + SIEM - Fleet Control - Compliance Packs - Open Source Ontology - Agent Identity - MCP Tunnels - Developer SDK - Multi-Agent Comms - Agent Marketplace - AI Cyber Insurance - Agentic Payments Network - Principal binding - Authorization - Data lineage - Policy - Decision record - PLATFORM - Own the LLM interaction chokepoint, own the product line ## 24. Vision - VISION - Vision - Arbiter of trust for all data - Every agent, every service, every device, every human action - arbitrated, logged, and explainable through a single gateway - Next generation of... - ServiceNow - $242B - Okta - $42B - Wiz - $32B - Proofpoint - $12B - 1Password - $6.8B ## 25. Problem option 1: mesh into a wall - People - 44 + i * 46; return ( - Review every one - Many agents - Many apps - ; if (k - ; return - Every agent × every app? - blocked - in review - People: one policy in - Wait for review - Agents stuck in pilot - The apps they need - 30 + i * 56; return ( - One hub, one policy - Agents: a mesh - Too many paths to review - Security is friction to agent adoption - Problem ## 26. Problem option 2: the review matrix - People - 44 + i * 46; return ( - Review every one - Many agents - Many apps - ; if (k - ; return - Every agent × every app? - blocked - in review - People: one policy in - Wait for review - Agents stuck in pilot - The apps they need - 30 + i * 56; return ( - One hub, one policy - Agents: a mesh - Too many paths to review - Security is friction to agent adoption - Problem ## 27. Problem option 3: the queue - People - 44 + i * 46; return ( - Review every one - Many agents - Many apps - ; if (k - ; return - Every agent × every app? - blocked - in review - People: one policy in - Wait for review - Agents stuck in pilot - The apps they need - 30 + i * 56; return ( - One hub, one policy - Agents: a mesh - Too many paths to review - Security is friction to agent adoption - Problem ## 28. Problem option 4: hub vs mesh - People - 44 + i * 46; return ( - Review every one - Many agents - Many apps - ; if (k - ; return - Every agent × every app? - blocked - in review - People: one policy in - Wait for review - Agents stuck in pilot - The apps they need - 30 + i * 56; return ( - One hub, one policy - Agents: a mesh - Too many paths to review - Security is friction to agent adoption - Problem ## 29. Solution option 1: one gate - 150 + i * 105; return ( - Support agents can't read the #finance channel on Slack - Your agents - Customer data - Customer's tools - Your agent - ✓ In production - 175 + i * 64; return ( - customer admin console · active policies - splunk> - ], ["Policies", - ], ["Leak prevention", - ], ["Any deployment", - ); return ( - SealGate: Govern how agents access data - ui-monospace, SFMono-Regular, Menlo, monospace - Sales agents only see Attio records for their own accounts - Block API keys and secrets in anything an agent sends - Log every agent action to Splunk - Q3 pipeline - #sales - Globex - key: sk_live_… - Acme - ALLOW - DENY - Audit logs - SIEM - RBAC - Policies - Leak prevention - SSO + SCIM - Tenants - Any deployment - Buried in security work? Hand it off. - Their IT sets policy like writing a prompt - SealGate handles the boring stuff - Solution ## 30. Solution option 2: policy in a prompt - 150 + i * 105; return ( - Support agents can't read the #finance channel on Slack - Your agents - Customer data - Customer's tools - Your agent - ✓ In production - 175 + i * 64; return ( - customer admin console · active policies - splunk> - ], ["Policies", - ], ["Leak prevention", - ], ["Any deployment", - ); return ( - SealGate: Govern how agents access data - ui-monospace, SFMono-Regular, Menlo, monospace - Sales agents only see Attio records for their own accounts - Block API keys and secrets in anything an agent sends - Log every agent action to Splunk - Q3 pipeline - #sales - Globex - key: sk_live_… - Acme - ALLOW - DENY - Audit logs - SIEM - RBAC - Policies - Leak prevention - SSO + SCIM - Tenants - Any deployment - Buried in security work? Hand it off. - Their IT sets policy like writing a prompt - SealGate handles the boring stuff - Solution ## 31. Solution option 3: spaghetti vs hub - 150 + i * 105; return ( - Support agents can't read the #finance channel on Slack - Your agents - Customer data - Customer's tools - Your agent - ✓ In production - 175 + i * 64; return ( - customer admin console · active policies - splunk> - ], ["Policies", - ], ["Leak prevention", - ], ["Any deployment", - ); return ( - SealGate: Govern how agents access data - ui-monospace, SFMono-Regular, Menlo, monospace - Sales agents only see Attio records for their own accounts - Block API keys and secrets in anything an agent sends - Log every agent action to Splunk - Q3 pipeline - #sales - Globex - key: sk_live_… - Acme - ALLOW - DENY - Audit logs - SIEM - RBAC - Policies - Leak prevention - SSO + SCIM - Tenants - Any deployment - Buried in security work? Hand it off. - Their IT sets policy like writing a prompt - SealGate handles the boring stuff - Solution ## 32. Solution option 4: the right keys - 150 + i * 105; return ( - Support agents can't read the #finance channel on Slack - Your agents - Customer data - Customer's tools - Your agent - ✓ In production - 175 + i * 64; return ( - customer admin console · active policies - splunk> - ], ["Policies", - ], ["Leak prevention", - ], ["Any deployment", - ); return ( - SealGate: Govern how agents access data - ui-monospace, SFMono-Regular, Menlo, monospace - Sales agents only see Attio records for their own accounts - Block API keys and secrets in anything an agent sends - Log every agent action to Splunk - Q3 pipeline - #sales - Globex - key: sk_live_… - Acme - ALLOW - DENY - Audit logs - SIEM - RBAC - Policies - Leak prevention - SSO + SCIM - Tenants - Any deployment - Buried in security work? Hand it off. - Their IT sets policy like writing a prompt - SealGate handles the boring stuff - Solution ## 33. Value prop option 1: the handoff - > never share customer data - ); return ( - vs - The builder - We ship sooner - The customer's security team - We stay in control - Agent in production - Production - Agents to production, faster with security - Single sign-on - Permissions - Audit logs - Leak prevention - Kill switch - With SealGate - Without - Weeks - Months - Approved - Blocked - Stuck in pilot - One integration - Built in-house, deal by deal - SealGate lane - Standard lane - Controls already in place - Security review queue - Value prop ## 34. Value prop option 2: the stack - > never share customer data - ); return ( - vs - The builder - We ship sooner - The customer's security team - We stay in control - Agent in production - Production - Agents to production, faster with security - Single sign-on - Permissions - Audit logs - Leak prevention - Kill switch - With SealGate - Without - Weeks - Months - Approved - Blocked - Stuck in pilot - One integration - Built in-house, deal by deal - SealGate lane - Standard lane - Controls already in place - Security review queue - Value prop ## 35. Value prop option 3: both sides win - > never share customer data - ); return ( - vs - The builder - We ship sooner - The customer's security team - We stay in control - Agent in production - Production - Agents to production, faster with security - Single sign-on - Permissions - Audit logs - Leak prevention - Kill switch - With SealGate - Without - Weeks - Months - Approved - Blocked - Stuck in pilot - One integration - Built in-house, deal by deal - SealGate lane - Standard lane - Controls already in place - Security review queue - Value prop ## 36. Value prop option 4: the checkpoint - > never share customer data - ); return ( - vs - The builder - We ship sooner - The customer's security team - We stay in control - Agent in production - Production - Agents to production, faster with security - Single sign-on - Permissions - Audit logs - Leak prevention - Kill switch - With SealGate - Without - Weeks - Months - Approved - Blocked - Stuck in pilot - One integration - Built in-house, deal by deal - SealGate lane - Standard lane - Controls already in place - Security review queue - Value prop ## 37. Appendix - Appendix ## 38. Moats (Seven Powers) - Moats - SealGate compounds through concrete defensibility - ); return ( - Cornered Resource & Scale Economies - Data flywheel - Finite Lloyd's coverholder slots and reinsurance capacity that early movers lock up - Full-stack verticalization from security to underwriting is difficult to replicate - flywheel - Switching Costs - Integration moat: policies, audit trails, approvals, and incident history become entrenched in SealGate. - Network Effects - Agent communication and data handoff protocols: more adopters → stronger standard → harder to bypass - Agent marketplace: network effects compound as SealGate becomes the standard for trusted agents - See appendix: Agent Marketplace - Salesforce - Slack - ServiceNow - Microsoft AD - Terraform - HashiCorp - Counter-Positioning - Enterprises want model interoperability and zero lock-in to any single lab. SealGate stays neutral across every lab, model, and data system - the control plane customers need to avoid vendor dependency. - Anthropic - OpenAI - Google DeepMind - logos - points - visual - Platform Data - Risk Scoring - Better Pricing - Lower Loss Ratios - More Capacity - More Customers - Agents - Identity - Apps - SIEM - Data stores - Policies Images: SealGate ## 39. Social: Guerrilla Marketing - Marketing - Security research is our distribution engine - Founder-led exploit demos of real AI products go viral, build credibility, and convert attention into inbound - at zero ad spend. - tweet - linkedin - On X · 1.5M views - Perplexity Comet's AI Agent just bankrupted your company - Eito Miyamura - On LinkedIn · 1,265 reactions - Clawdbot just injected malware into your code - Eito Miyamura - On LinkedIn · 928 reactions - Break real AI products - Distribute organically - Convert to inbound - Position SealGate as the solution - dark ## 40. Agent Marketplace - Appendix - Agent Marketplace - Since SealGate manages the keys to the kingdom - the company context third-party agents need to be useful - the platform can expand to become the arbiter of all external agent communication. - SealGate handles - Context + RBAC - SealGate Gateway - Arbitrates trusted context for agents - Marketplace - Apps - Identity - Docs - Tickets - Logs - Data - RBAC - allow - deny ## 41. Agent Runtime Control - Appendix - Agent Runtime Control - Agent Runtime - Policy - Memory - Tools - Logs - Owner - Authority - Version - RBAC - TTL - Task - Handoff - Prompts - Calls - Data - Approvals - Identity - Runtime - Permissions - E-Stop - Approval - Kill - Forensics - Activity Trail ## 42. Incumbent Acquisitions - Appendix - Incumbents are buying, not building - $6B+ across 20 bolt-on acquisitions - no single incumbent covers all 7 capability areas - Incumbent - Acquisitions - Spend - † Announced, pending close - Supply Chain - Prompt Security - Shadow AI - Red Teaming - AI Identity - AI Governance - MCP Security - $3.3T - Built internally - Defender AI-SPM - Prompt Shields - Purview AI Hub - PyRIT - Entra Agent ID - Purview - $475B - Robust Intelligence - Astrix - Galileo - $800M+ - $160B - Protect AI - Koi - Portkey† - $1B+ - Portkey - $165B - Pangea - SGNL - Adaptive Shield - Flow Security - $1.5B+ - $100B - Veza - data.world - $1.4B+ - $28B - SPLXAI - Avalor - $350M+ - $22B - CalypsoAI - Heyhack - $180M+ - $13B - Lakera - Cyata - Atmosec - $450M+ - Private - Aim Security ## 43. Cyber Insurance Market - APPENDIX - Cyber insurance is becoming a security distribution channel - Why it matters - Insurers increasingly price cyber risk based on controls. - SealGate can become the control evidence layer that helps businesses qualify for coverage, reduce risk, and unlock AI rollout. - Sources: Munich Re, Swiss Re - 2025 premiums - ~$15B - 2030 premiums - ~$28B - Growth - ~2x ## 44. Agent Insurance - AGENT INSURANCE - Sell the "seat belt", underwrite the policy - The "seat belt" - SealGate Platform - Recurring per-seat SaaS subscription for AI agent security & data control - ACV - $80k–$150k - ENABLES - DISTRIBUTES - The "car insurance" - AI Cyber Insurance - Recurring revenue share with insurance partners; platform data enables better risk underwriting. Partnering with a large insurer with 6M+ policyholders. - seat belt - car insurance ## 45. Further Reading - Appendix - Further Reading - blog - notebook - Agentic AI Disrupts Traditional Data Security - 3 min - Insurance Industry - 8 min - Lloyd's SIAB - 34 min - Cyber Insurance Industry - 33 min - Interactive visual - Blog post Sources: https://edisonwatch.substack.com/p/agentic-ai-disrupts-traditional-data , https://eito.me/notebook-exports/insurance_industry.htm , https://eito.me/notebook-exports/lloyds_siab.htm , https://eito.me/notebook-exports/cyber_insurance_industry.htm ## 46. FAQ - Appendix - FAQ - Why not go all-in on insurance? - The insurance industry doesn't move at AI speed. Heavy bundling dynamics and slow structural change make it challenging to go all-in on today - but we're positioned to capture structural shifts as they emerge. ## 47. RBAC - THREAT - AI Agents Break RBAC - Agents forward data across permission boundaries. Existing controls can't see it. ## 48. Prompt Injection - THREAT - The Lethal Trifecta - AI agents with data access + untrusted content + external comms = data exfiltration ## 49. Org MCP Push - SOLUTION - Org-Wide MCP Deployment - Admins toggle approved MCP servers in the gateway. Every employee laptop gets them instantly. ## 50. Trifecta Defense - SOLUTION - Trifecta Defense - Without SealGate, a poisoned email exfiltrates data. With SealGate, the leak is blocked. ## 51. User Accountability - SOLUTION - User Accountability - Auto-deny risky actions or route approvals to the responsible human. Every agent action has an accountable owner. ## 52. Quarantine - SOLUTION - Automatic Quarantine - Rogue MCP servers detected and quarantined before they can exfiltrate data ## 53. Stdio vs HTTP - APPENDIX - STDIO vs HTTP - Local STDIO execution gives admins zero visibility. Remote HTTP through a gateway enables full governance. ## 54. Code Mode - SOLUTION - Code Mode Security - Three defence layers protect MCP-as-code: AST analysis, Deno sandbox, and taint-aware trifecta. ## 55. Desktop Install - SOLUTION - Frictionless Desktop Install - One-click install, guided setup wizard. Agents are protected in under a minute. ## 56. Agent Bypass - PROBLEM - Agentic AI breaks traditional data security - AI agents are often overprivileged, sprawling everywhere (shadow AI/Connectors), no ownership, no activity trail, no unified controls - creating blind spots and risk. ## 57. Proxy - SOLUTION - SealGate controls how agents access your data - Secures and unifies agent data access. Stop data leaks before they happen for any agent, on any device. ## 58. Scale Pilot - Enable AI to do real work, without risk - Agent interoperability: avoid lock-in - Unified visibility and control - VALUE PROP - For CISO/IT Admins to turn security into business value