Your people can use any AI on your cloud, code and chat. Nothing it does can leak your data or break production without a named human saying yes, and you can prove it.
Sunkar is the identity and trust layer between AI agents and everything they touch.
Requested
2026-11-12 · 11:53:32 UTC
Approved3.3 s later
11:53:36 UTC
- Class
- change.safe
- Provider
- AWS
- Run
- dry run first
Numbers from a paying customer's account.
68
17regions scanned
Unrestricted outbound traffic to the internet
11Object storage access logging disabled
11Object store missing MFA-delete protection
11Security group open to the world
9Block volume is not encrypted at rest
5Compute instance has a public IP
5
6
Requests
6
Decided by policy
2
Approved by a person
4
Executed
Injection attempts flagged
2
How one leaked key and one AI agent become a breach.
Path open · 3 of 3 steps reached
Claude Code agent with that key. The agent runs with that key. Nobody is asked, and CloudTrail records the human, not the agent.
Five steps before an agent touches anything.
- 01
Identity
- 02
Policy
- 03
Approval
- 04
Execute
- 05
Ledger
Results are cleaned before they reach the model.
Keys and tokens
Personal data
Injection, in Arabic too
Install once, in the tools your team already uses.
- Claude Code
- Claude Desktop
- Cursor
- GitHub Actions
- Amazon Bedrock
- AWS
- Kubernetesnext
- GitHubnext
- Any identity provider, via SSO
# Claude Code claude mcp add sunkar -- npx -y @sunkar/gateway
Free: one agent, read-only. On npm as @sunkar/gateway.
Sovereignty and regulation
- Hosted in Doha
- In-Kingdom option (Sovereign)
- NCA ECC-2
- PDPL
- SDAIA AI principles
- NCA AI guidelines (draft)
Mapped to these frameworks, not certified by them.
Live with paying customers in KSA and the UAE. Hosted in Doha today, in-Kingdom option for Sovereign.