Open source / identity

Ory is having an Identity crisis

On release cadences, security fixes, and what 'open source' is supposed to mean.

Why I really liked Ory Hydra

Ory Hydra is a genuinely well-designed piece of software. Its best idea is the login and consent flow. Hydra is an OAuth 2.0 and OpenID Connect server that deliberately doesn't hold a database of users. Instead, it redirects the browser to a login and consent app that you write, hands it a challenge, and waits for your app to accept or reject the request through Hydra's admin API.

That separation is what makes Hydra so flexible. Your identity backend can be LDAP, an existing user table from a legacy app, a hardware token check, or something nobody at Ory has thought of. Hydra does the hard, standards-heavy OAuth2 work, and you keep full control of who your users are and how they prove it, because Hydra is designed to 'hand off' to your dedicated Login and Consent application (LaCa) for that actual credential check. A lot of identity products force you into their user model. Hydra doesn't, and that's a design choice I've always admired.

It was also crucial for an organisation I contracted to, who was moving from SAML to OIDC and whose own IDP API was already well-established, mature, and agnostic to the actual auth protocol in use. We had been considering other options such as Gluu at the time, but couldn't find anything quite like what Hydra allowed with its LaCa.

Which is why the announcement I'm about to discuss is so disappointing.

The announcement

On 11 November, Ory's founder and CTO published "Doubling Down on Open Source: Predictability, Security, and a New Release Rhythm" . From the headline you'd think Ory had decided to invest more in Hydra, Kratos, Keto and Oathkeeper, the open source stack it built its name on.

I read the post and I found the complete opposite.

Day-to-day development is no longer public.

Timely security fixes for self-hosters are now a paid product.

And the free edition is now positioned for evaluation and prototyping - where you can perhaps live with waiting for fixes.

The headline says "doubling down". The substance, as I read it, says "pay up". Or "doubling down" in the negative sense, especially given that Ory has been pressed for clarity on their security lifecycle again and again .

What was actually announced

Strip out the framing and the policy, is this:

  • Ory Open Source moves from roughly one release a year to two, starting in 2027, with no committed dates. (On Github, they say this is what the 'doubling' refers to.)
  • Since the end of July 2026, the public repositories no longer receive code commit by commit. They get it in one go with each release.
  • Apache 2.0 users get security fixes when a release ships. There's no security SLA, no guaranteed date and no out-of-band patch releases, and fixes go into the latest release only.
  • A new commercial product, the Continuous Security License (CSL), provides patched builds of self-hosted Open Source as vulnerabilities are resolved. It's priced per component, and the price isn't published: you have to contact Ory to get a quote.
  • Paying customers get those patched builds without the source of the fix. The source, fix and advisory become public with the next open source release.

Two releases a year is presented as the 'good' news. To me it's the sweetener on the real change: the open source project now trails the product Ory's customers run, and closing that gap is something you have to buy.

Doubling down on what, exactly?

Code developed out of public view and published on a schedule still carries Apache 2.0, and that matters. But a lot of what makes open source work as a practice goes with it: public history, reviewable changes, outsiders following along, catching regressions, contributing, and forking with confidence.

Think about what a would-be contributor also now faces. The public branch could be months behind Ory's internal tree. Any pull request you write targets code that may already have been refactored, renamed or fixed internally. You can't see what's in flight, and you can't review what's coming until it has shipped.

Ory explains that it holds back all changes, not just security fixes, partly so that the public history doesn't point at where the security work is happening. I understand the reasoning. But the effect is that the public record of the project is now intentionally less informative. That's a strange thing to call "doubling down on open source".

There's some history here, too. Ory's own page on its commitment to open source says that commitment "isn't a marketing ploy." Its Enterprise License page warns that "waiting on the community to address issues may result in delays and fines".

What I think Ory fails to read from the room is that they spent years building that community, as all good open source projects do. They're now warning you about relying on it.

"They just aren't free"

The line in the post that sums it up best is Ory's own. The fixes, the post says, are available right away. "They just aren't free."

Read that as someone running Ory in production. This is your identity layer: logins, sessions, OAuth tokens. When Ory knows about a vulnerability and has a fix, you can buy that fix now, or wait for (and gamble your chances on waiting for) the next open source release, which under the new policy has no committed date.

It's worth thinking about where vulnerability reports come from, too. Ory runs a disclosure process and a private bug bounty, and reports can come from anyone, including people in the open source community. Under the new policy, the fixes for those reports reach paying customers first, and open source users last.

I don't accept the implied message that if you really cared about security, you'd be paying. Making Apache 2.0 users wait is a choice Ory has made. The way I see it, the CSL isn't really charging for the work of fixing, because that work gets done for paying customers either way. It's charging for not having to wait.

None of this denies that sustaining an open source company is hard, or that security work costs money. Ory says that without this revenue there would be less development, and I take that at face value. But there are ways to charge for security work that don't leave free users knowingly behind, and I think Ory has picked the wrong one.

You know what this sort of pricing model reminds me of? Selling vulnerabilities to the highest bidder. The fact that this new policy attracts those who 'surely would pay for it if they wanted it that badly' in a similar fashion, is an ugly look.

Apache 2.0 on the label, proprietary licence in the box

Ory is careful to repeat that every release is published in full under Apache 2.0. That's true. But consider which version of Ory is current and fully patched at any given moment: the builds running on Ory Network and supplied to Enterprise License and CSL customers. All of those are commercial.

Per Ory's own docs, enterprise builds aren't available for public download and need a licence agreement to run. CSL customers get patched builds without the source of the fix. So the most up-to-date Ory is a set of binaries under a commercial licence, and the Apache 2.0 Ory is a periodic snapshot of where that product was some months earlier.

The CSL also has no published price. You're told the responsible way to run Ory in production is to buy a licence whose cost you only learn by contacting sales. For security infrastructure, where knowing exactly what you run and what it will cost is part of the job, I find that a poor fit with the openness that drew people to Ory in the first place.

Even the safety valve that makes permissive licences meaningful is weakened. You can still fork, but only from a snapshot that may be months behind, without the commit history that explains how it got there.

We've seen companies like HashiCorp, Elastic and Redis change their licences outright in recent years, and take the reputational hit that came with it. Ory has kept its licence. Instead, it has moved the most current, patched version of its product behind a commercial agreement, while the Apache 2.0 label stays on the public releases.

It then presents the decision as a false dilemma : "The alternative was moving to a non-free license.".

What doing it properly looks like

If the goal is security and sustainability, there are well-trodden paths, none of which ration fixes to users at the same time as telling them that AI threats against them are growing:

  • Develop in the open, as Ory did until July. Keep public history public. Bonus: trust in your project grows with you.
  • Use coordinated disclosure. Fix privately, and perhaps give downstreams and paying customers a short embargo to prepare themselves, then release to everyone at once. Some customers will pay for the advance notice to mobilise their team or mitigations.
  • Sell what's actually scarce: long-term support and backports for older versions, 24/7 support, zero-downtime migrations, compliance paperwork, hosting, enterprise features. Ory already actually sells most of these. I note that they count OpenAI as a customer . Now, OpenAI's business model might be losing money, but I bet they pay Ory alright.
  • Publish your prices. If a commercial licence is part of the deal, let people see what it costs. If you aren't going to change your decision, then transparency is better than nothing at all.

None of these require telling people who believe in your product that security is a premium.

Where self-hosters can look instead

Ory's post is framed as a commitment to open source, but is structured, in my view, like a sales funnel. The cadence change is real but minor. Their own point is that AI-driven security threats are growing: I don't honestly believe that 'twice a year' is a well-intentioned answer to that threat. The substantive change is that public development has stopped and timely security fixes for the free edition are now a product.

Ory's own edition guidance says the free version is fine for production where waiting for fixes is acceptable. For an identity provider, I'd argue that's rarely acceptable. I think the mistake is also in assuming that good Identity Management is an ambition solely deserved by corporations with money. Identity, after all, is about the Individual, and that means human beings. There are community organisations, non-profits, and other bodies who deserve security around the identity of everyday people. These are also the communities that see their values reflected in open source and what drove them to products like Hydra in the first place. I know this because I have spent decades working with such communities.

So the real choice I see that Ory has put to self-hosters could be considered: pay up, or plan your exit.

Plenty will choose the second.

So what does an exit look like?

None of these are drop-in replacements for Hydra's login and consent design, which I will sorely miss. And each makes different trade-offs. But all of them ship security fixes to their free users:

  • Authentik is a full identity provider (OIDC, SAML, LDAP, SCIM, RADIUS and more). Its core is MIT-licensed, and Authentik Security sells an Enterprise edition with extra features and support. Crucially, that Enterprise edition uses the same container images as the open source version : you unlock it with a licence key. Its security policy releases fixed versions for all supported releases, after a public announcement at least 24 hours ahead. Paying for Enterprise doesn't get you security fixes that the free edition is denied. That's open core done in a way I can respect.
  • Authelia is a lightweight authentication and authorisation server written in Go, offering single sign-on, two-factor authentication and an OpenID Connect provider. It's Apache 2.0, with no paid edition, and is funded through sponsorship. Its security policy follows coordinated disclosure, and fixes are released to everyone. Notably, Authelia handles the "attackers read the advisory" concern by publishing a brief initial advisory alongside the fix, then adding the fuller details roughly a week later. It delays the details, not the fix.
  • Keycloak is the heavyweight option: Apache 2.0, a Cloud Native Computing Foundation project, and widely deployed. Red Hat sells a supported build, but the upstream project is developed in public and is where fixes land.

If you're running Ory self-hosted today, I'd read Ory's post carefully, work out what a six-month wait for a fix would mean for your users, and then decide whether the CSL, or one of the projects above, is the better answer.

Weighing up your identity stack?
I help organisations run and secure self-hosted infrastructure, including identity providers. Get in touch if you'd like a second opinion.
Did you appreciate this article? Any support is appreciated!