Insights

AI Security

Which of your endpoints actually need to be public? An afternoon exposure review

JPMorgan Chase's CISO is telling security teams to stop letting anonymous traffic reach their services. For a HealthTech company, the first step is an afternoon spent listing every public endpoint and asking who is supposed to use it.

November 11, 2026 · 6 min read

Which of your endpoints actually need to be public? An afternoon exposure review

Every endpoint that answers an anonymous request is one an automated agent can probe for free. On November 7, Pat Opet, whom ctxsec identifies as JPMorgan Chase's CISO, put it bluntly on X: "if you're working for a company in cybersecurity, you should be evaluating your options to abstract your companies services from the internet." His reasoning was one line: "why allow an unauthenticated anonymous agent of an attacker to interact with your systems?"

The post traveled fast and drew pushback. Some readers heard "take the bank offline." That is not what he said, and the follow-ups matter more than the headline. For a HealthTech company with a small engineering team, the practical version of his point is an exposure review you can finish in an afternoon.

What Opet said, and what he did not

In a reply the same evening, Opet wrote that the post "says nothing about being closed nor customers agents, it's about maximizing the use of authentication and authorization to strengthen your perimeter." In another, he described one way to get there: an identity-aware proxy where "identity includes customers" and "all anonymous traffic from the internet is dropped before it reaches services," adding that it is "just one way to achieve the objective."

He also described what happens after sign-in: if you segment at the identity layer, "an authenticated attacker is only authorized to see specific pages/functions/endpoints," and when abuse is detected, "it's an automated account suspension with a total failure blast radius of one."

So the idea has three parts. Anonymous traffic stops at the edge. Signed-in users see only what their role needs. Abuse costs the attacker one account, not your whole application. None of that is new architecture. What has changed is the cost of probing. An unauthenticated endpoint used to wait for a person to find it. Now anything that can run an agent can enumerate and test it all day.

Fintech is one of the regulated industries I have worked in, and the habit I took from it is asking why a service is public before asking how to defend it. Healthcare vendors often skip that first question, because their products grew up as web apps.

Why this lands hard in HealthTech

A typical HealthTech product has more public surface than its team thinks. The marketing site and the login page are meant to be public. Then there is everything else that ended up with a public DNS name because it was convenient: a staging environment, an admin panel, an API used by one integration partner, a webhook receiver, a FHIR endpoint for a pilot, a support tool, a metrics dashboard someone shared during an incident.

Each of those answers anonymous requests. Each has its own authentication, written at a different time by a different person. A flaw in any one of them is reachable by anyone. Opet's framing flips the default: a flaw in the application should not be enough on its own, because an anonymous caller should never reach the application in the first place.

The afternoon review

This is a list, not a project. Block four hours and do it with whoever knows your DNS and your cloud accounts.

  1. List every public hostname. Pull it from three places, because each one misses things: your DNS zones, your cloud provider's load balancers and public IPs, and certificate transparency logs for your domains. Certificate logs are good at surfacing the staging host nobody remembers creating.

  2. For each hostname, write down who is supposed to use it. Use five buckets: anyone on the internet, signed-in customers, your staff, a specific partner or integration, or nobody you can name. Be honest. "A partner we onboarded last year" is a real answer and an important one.

  3. Check what an anonymous request gets. Open each one in a private browser window or with a simple request from outside your network. Does it return a login page, a public page, an API error, or real data? Do not test anything beyond your own systems, and do not try to bypass anything. You only want to know what the front door shows.

  4. Mark every mismatch. Anything in the staff, partner, or "nobody" buckets that answers anonymous requests with more than a login prompt is a finding. So is anything in the customer bucket where the application, not the edge, is the only thing checking identity.

  5. Assign an owner and a decision to each finding. For each one, the decision is one of four: put it behind an identity-aware proxy or SSO at the edge, restrict it to known partner connections such as mutual TLS or an IP allowlist, move it to a private network, or shut it down.

At the end you have a short table: hostname, intended users, what anonymous traffic sees, owner, decision, date. That table is the deliverable.

What usually comes out of it

The first pass almost always finds the same three things. Staging and test environments with production-like data and a public address. Admin and support tools protected only by the application's own login. Partner APIs that accept traffic from anywhere because restricting them meant a conversation with the partner.

The fixes are not exotic. Staff tools go behind the same SSO you already use for email. Partner APIs get mutual TLS or an allowlist. Staging goes private or gets synthetic data. The customer-facing app keeps its public login page, but you look at whether anything behind it is reachable before sign-in.

The harder change is Opet's second point, segmenting after authentication so a signed-in attacker sees only what their role allows. That is design work, and it will not fit in an afternoon. But you cannot plan it until you know which endpoints exist and who they are for.

Questions a health system will start asking

Health system security reviewers already ask about MFA and encryption. Expect them to ask a version of Opet's question as automated probing gets cheaper: which of your services can an anonymous caller reach, and why? A vendor that can hand over the table from this review, with owners and dates, answers that in one page.

Run the review once, then repeat it every quarter, because public endpoints come back. Someone will spin up a demo host before a sales call or open a webhook for a new integration. The quarterly pass catches it. If you ran it this week, how many rows would land in the "nobody you can name" bucket?