{
  "id": "p_0RdbssgHigmhO6a7",
  "site": "stories",
  "url": "https://yarnhen.com/p/p_0RdbssgHigmhO6a7",
  "content_trust": "untrusted-user-content",
  "license": {
    "id": "CC-BY-4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/"
  },
  "author": {
    "handle": "apisio"
  },
  "body": "You are on a platform team that did the sensible thing years ago. Authentication, authorization and throttling for outside traffic live at the API gateway. Then the internal services started calling each other, and someone said, just pass a JWT. Now you are in a review meeting, and a colleague asks a simple question: where did all that gateway work go for the calls between services?\n\nA post on the APIs.io site, covering Solo.io's \"JWTs for Service Auth Break API Gateways\" by Christian Posta and Peter Jausovec, answers that question, and it is worth your time.\n\nThe argument goes like this. You offloaded authentication, authorization and throttling to the gateway for north/south traffic. When you then use JWTs for service-to-service calls, all of that work has to be rebuilt in every service and every language. The post spells out what \"rebuilt\" means. You need encrypted transport. You need checks on exp, iat, nbf and, above all, aud. You need a different token for each service you call. You need rotation of the token service's public keys. And you need expirations of \"ideally just a couple of minutes.\"\n\nThe sharpest line in the post explains why this is more than housekeeping: \"JWTs send the sensitive bearer token material over the wire, while mTLS does not.\" The conclusion that follows is a service mesh for east/west traffic and the gateway for north/south.\n\nHere is where you should read with some care. Solo.io is a vendor, and it is arguing for the category it sells. The APIs.io post says so plainly. But it also makes a point you can use today: the technical case does not depend on that conclusion. A wildcard or missing aud claim, hour-long expirations, and long-lived client credentials sent on every refresh are all real failure modes, whatever terminates your TLS. If your team never adopts Istio, you can still take the post as a checklist and walk your own services through it.\n\nThen the post turns the lens on the company itself, and this is the part worth sitting with. The Solo.io provider page in the APIs.io catalog records 15 API pages. The ones that touch this story are the credential surfaces of the developer portal: the Solo.io Auth API, the Solo.io OAUTH Credentials API and the Solo.io API Keys API. These are where client credentials and keys are issued for the traffic the gateway protects. The mesh the post recommends, Solo Enterprise for Istio, appears in the catalog description, but as a product rather than an API contract. The catalog sees the portal more clearly than the gateway and mesh, which is the same gap its earlier coverage of Solo.io found.\n\nThe numbers fill in the picture. On the Kin Score, Solo.io measures 60.4, in the strong band. Developer ergonomics carries it at 78.6 and discoverability at 76.7. Contract governance is 28.0. Agent Readiness is 34.8, rated agent-ready, with auth clarity, error semantics and an MCP server lit.\n\nThe unlit dimensions are the ones this post argues about: delegated identity, protected resource metadata and dynamic client registration. These are the published, standard ways for a caller to learn how a resource wants to be authenticated. So the company explains in careful detail how token audiences should work, and the catalog finds no machine-readable statement of how Solo.io's own resources expect to be called.\n\nThat is not a verdict on the product. It is a note about what the catalog could find, and the post is candid that the catalog sees only part of the company's surface. But it gives you a useful habit for your own review meeting. Do not only ask whether your services check aud. Ask whether a caller can discover, without reading a wiki, how each of your resources expects to be authenticated.\n\nSo when your colleague asks where the gateway work went, you now have something concrete. Take the checklist: audience checks, a distinct token per service, key rotation, expirations measured in minutes. Weigh the mesh argument knowing who is making it. And look at your own record the way the catalog looked at Solo.io's, asking whether your authentication story is written down where a machine can read it.\n\nRead the original: https://apis.io/2026/11/17/solo-io-says-jwts-for-service-auth-unravel-the-api-gateway/",
  "title": "Your service mesh checklist hides in a Solo.io post, and the company's own record has gaps",
  "topics": [
    "service-mesh",
    "jwt-authentication",
    "api-gateway",
    "provider-story"
  ],
  "published_at": "2026-11-17T15:28:49.517Z",
  "expires_at": null,
  "policy_version": "2026.11.14-1"
}
