This is a big one. A massive one. It's the culmination of a solid 7 months of work that finally, as of now, is live. The full back story is in my blog post from mid-June about The Big 5 Announcements but to save you trawling through all of that, here are the cliff notes:

1. Domain searches in HIBP are resource intensive and the impact was becoming increasingly obvious
2. More than half the Fortune 500 are using this feature, along with a who's who of big brands
3. We decided to introduce pricing tiers to the largest domain searches...
4. ...but also add stuff, most notably domain searches by API and formal support...
5. ...and remove stuff, most notably the need for verifying control of a domain after you've done it once

I've spent the last 8 weeks since publishing that post crunching numbers, writing code, doing loads of formal things (namely terms of use and privacy policy), and regularly talking about it on my weekly video. I've had loads of enormously useful feedback, much of which has shaped the state of the services we're launching here today. Thank you everyone who contributed, now let me get into it and explain exactly what we've come up with 🙂

The Pricing Structure

We've been thinking about the best way to structure this since January. How do we take something that has been provided for free for almost a decade and put a reasonable price on it? That's a highly subjective word - reasonable - and there'll never be complete consensus, so it's more about passing the pub test where your average person will look at this and go "yeah, that seems fair enough". Let me explain the thinking and how we reached the pricing structure you'll see further down:

Firstly, we wanted most domain searches to remain free. This keeps with the spirit of HIBP's roots being a community service and ensures the data is accessible without barrier to the majority of people. It would also mean that for most people, these changes would have absolutely no impact on the way they've been using the service, not unless they want access to the new bits.

Next, we wanted to divide the commercial offerings into a manageable number of tiers. The public API key has 4 tiers and I reckon that's the sweet spot; it's not too many options, but it's enough to provide a good separation between the scale of each. We then wanted to distribute the number of domains that would fall into the commercial category roughly equally between those 4 tiers, so it was pretty much a matter of taking what was left after the free ones and dividing them into 4 groups and putting a price on them.

Finally, we wanted the first commercial tier to be easily affordable so that most people could access it without thinking twice about it. My measure for that has always been "the cost of a cup of coffee", so I went down to my favourite local and checked what I was blindly paying when I waved my watch in the general direction of the EFTPOS machine:

$6 Aussie, or just under $4 in USD. Which led us to here (all in USD from now on):

What you're looking at here is a list of plan names (more on that soon), the size of the domain it covers (expressed in the number of breached email addresses on it), what percentage of all domains presently being monitored in HIBP this represents and, of course, the monthly price. As with the public API, if you subscribe annually then it's "pay for 10, get 12" which means that "Pwned 1" price works out at only $3.25 a month. As I flagged in the earlier post, this is all based around the number of addresses that appear in a breach, with one important caveat I'll expand on later: this number excludes all breaches flagged as a spam list. As a rough rule of thumb, over the years I've found approximately 20% of addresses on a domain have been breached so by that logic, you'll need 55 actual email addresses on a domain before there's a cost. Or up to 130 before it costs more than a coffee a month. (If you're a stickler for detail and are thinking those percentages are too perfect, I've rounded them from their actual values of 59.1%, 9.7%, 11.3%, 10.4% and 9.4%.)

But what if you have multiple domains? Easy - the one plan will cover all your domains within the size of that plan. For example, if you have 3 domains and one has 5 breached addresses, one has 20 and one has 90, you can get a single "Pwned 2" plan and cover them all. Or get a single "Pwned 1" plan and cover just the first 2. It's pretty simple.

So that was our initial thinking - stand this up as a product that sits alongside the existing API key one then you just purchase whichever one you want. Then, Brendan gave me a much better idea - combine them altogether! You can see the gears turning around in my head as I read his suggestion and as the days progressed and I gave it more thought, it became a brilliant idea. It massively simplifies the code base, it removes a lot of confusion that I'm sure would have otherwise ensued and perhaps most importantly, it gives you all something more than you would have had otherwise. The one fly in the ointment was the price disparity; the above prices are 13% to 15% higher than the old corresponding API key ones. So, what we've decided to do is run the old prices until 8 October then revise everything to the new prices above. That gives more than 60 days' notice to everyone with an existing API key (we'll have to email everyone anyway as the terms of use have changed to incorporate the domain bits), and there's clear verbiage everywhere about the change for anyone purchasing a new subscription. Plus, it gives everyone a little incentive to lock in for a year now and delay the increase until later in 2024. Thanks Brendan! 😊

So that's the rationale. There's no change for 60% of domains that have previously been searched, a negligible cost for the next 10% of them with the remainder paying commensurately more based on their scale. But we didn't just want to whack a cost on an existing service and you're down a few bucks a month with nothing more to show for it, let's talk about new stuff!

But Wait, There's More!

There are two brand new features we're now offering to all commercial subscribers. Even if your domain is small and has less than 10 breached addresses on it, you can still get access to these features via the entry level plan and they're both pretty self-explanatory: API-level access and formal support.

API first as I think it's the coolest and it's exactly what it sounds like: there's now a public endpoint you can throw a domain at and get a JSON response of breached aliases and the incidents they've appeared in. It looks just like this:

GET https://haveibeenpwned.com/api/v3/breacheddomain/{domain}
hibp-api-key: [your key]

Which then responds like this:

{
  "alias1": [
   "Adobe"
  ],
  "alias2": [
    "Adobe",
    "Gawker",
    "Stratfor"
  ],
  "alias3": [
    "AshleyMadison"
  ]
}

If you're already paying for an API key, you have immediate access to this! Same key, same logic in terms of resolving the returned breach name to the full thing via the unauthenticated API that returns breach metadata, the only caveats are that is has to be a domain you've previously demonstrated you control and it has to be within your plan size (e.g. you have a Pwned 1 plan and your domains don't exceed 25 breached addresses). Otherwise:

Subscription upgrade required.

Just one more thing with the domain search API: it only makes sense to hit it after a new breach is loaded. There's absolutely no point in hammering away at it non-stop as you'll only get the same result so instead, try polling the brand new API we've just added to return only the most recent breach (it's massively cached at Cloudflare anyway) and just hit the domain search API when there's a new one. But because not everybody will do this and domain searches are expensive relative to other queries, the terms and conditions include this clause:

Controls such as rate limiting may be added to the domain search API if excessive API requests are made despite no new breaches appearing since the last request.

There is a rate limit based on a variety of factors and it's possible you may receive an HTTP 429 if you request it more frequently than is necessary. The only reason I'm not going into the details of how that works here is that I expect it will adapt and change pretty frequently in response to how people use the service. What I can confidently say now though, is that if you use the domain search feature in the way it's intended to work - querying each domain after a new breach is added - you won't have a problem with rate limits.

I'm really excited to see how people will integrate this data into their existing tooling, do please let me know if you do something awesome 😊

Then there's the formal support which we offer via Zendesk at support.haveibeenpwned.com. That launched with the API key upgrades last November and since that time, we've answered almost 600 tickets. We've been trying to fine tune things to the extent that the knowledge base there answers the most common questions, but there's certainly a great deal of time that still goes into supporting the questions that pop up. Adding domain searches to the mix will inevitably increase that, possibly by a significant order of magnitude which is why we're only making this available to commercial subscribers.

So, that's the new bits. If you're in that 60% group of people with smaller domains outside of the commercial tiers, you can get access to both the API and support by subscribing to the smallest possible plan for that cup of coffee a month. We feel that's a pretty reasonable balance, and I hope you do too.

Speaking of reasonable, about those spam lists...

Data Breaches Ain't Data Breaches

I mentioned sharing as much as I could in my weekly update videos, including the intended pricing structure and how it would be based on the number of breached email addresses on a domain. Several people raised a very important point as it related to the calculations: data breaches ain't data breaches or more specifically, there are breaches in HIBP that shouldn't be treated like the other ones as they artificially inflate the pwn count. Could these be excluded?

The Onliner Spambot incident was the worst culprit and in the case of one person that contacted me, it caused his personal domain to read as though hundreds of addresses had been breached when the correct number was... zero. Someone else had their domain pegged at 40 breached addresses whereas once you took this breach out, the number came down to 13. This created somewhat of a rock and hard place situation because whilst those aliases did appear in this incident, they weren't real addresses. But what's a "real" email address anyway? Or more specifically, how can I tell via a string alone whether an address is real or not? A decade ago now I wrote about how hard this is and per the comments on that post, concluded that the only way to tell for sure is to send an email and have the recipient perform some sort of explicit action such as clicking on a link. Clearly, that's not feasible in this situation but equally, putting a price on a service based on a metric that has been artificially inflated just wasn't fair.

Adding spam lists back in 2016 was the right thing to do but equally, excluding them from the number that determines the pricing tier is also the right thing to do. We've tried to make this logic as clear as possible throughout the system and focus on a simple UX that's explicit but can also provide more insight if required,

And if you're interested in which breaches specifically have been classified as a spam list, I've added a filter to the API that lists all breaches. It's an unauthenticated API you can load directly in your browser via GET request and at the time of writing, has 11 breaches on it with nearly 1.4 billion records.

The very last thing from that screen cap is the "Enable debug mode" link and for that, we need to talk about "domain creep".

Domain Creep, and Getting What You Paid For

Data breaches are obviously an ongoing thing. Always have been, always will be so what that means is when you look at a domain today and see, say, 20 breached accounts on it, that might be 30 breached accounts tomorrow. I think everyone who uses HIBP understands that, but it does create a bit of a problem when domain searches are priced on a metric that can "creep". What if you've just paid for a year's worth of Pwned 1 subscription and per the example here, you've suddenly got more than 25 breached accounts on your domain and can no longer search it?

The sentiment of how this should be handled was always obvious: people have to get what they pay for. We didn't want a situation where someone could be left disappointed, and our fear was that the organic increase in breaches could lead to that event. The solution was easy: when you buy a subscription at a certain scale, every domain you're currently monitoring that can be searched on the first day of the subscription can still be searched on the last day of the subscription. If you take out one year of Pwned 1 today and per the example above, the domain creeps beyond 25 breached accounts tomorrow, it'll have zero impact for the next 364 days.

I'm conscious that this concept can get confusing: domain searches are based on the number of breached accounts on the domain but not including spam lists and then locked in at the size of the domain until the next subscription renew... phew! The debug mode link mentioned above aims to show all this logic in its raw detail:

Even though domain1.com in this example has grown to 26 breached addresses, because it was 22 breached addresses when the subscription was taken out then that's the number it's locked at until it renews in August next year. I hope this is clear enough, do please leave a comment if we can do better.

Lastly, let me put some raw numbers around the "domain creep" situation as I foresee this causing concern beyond what might be warranted. Let's start with the number of unique email addresses which is approximately 6 billion. There have been about 723M records added in the last 12 months and a bunch of those will be for the same email address (shout out to everyone who was pwned again in the last year!) Further, of that number, most email addresses were already pwned. That's a link through to the Twitter feed where I broadcast the percentage of previously seen addresses and you'll see that number is regularly around the 60% to 70% range. In other words, it's probably in the order of 250M new addresses we've seen in the last year which is appx 4% of the entire corpus. So, yes, over the course of time we'll see domains slip into higher plans, but only at about the rate of CPI.

Lastly, locking domain counts for the duration of the subscription creates additional incentive to make it an annual one, and that's beyond the existing incentive of "buy 10 months, get 12 months". That's also in addition to massively cutting down on the number of times you may need to deal with corporate bureaucracy. Speaking of which...

Satisfying Corporate Bureaucracy

Let me start with a story: Many years ago during my lengthy tenure at Pfizer, I pushed hard to drive us away from traditional hosting models and towards modern cloud paradigms, namely the Azure App Service. Here we had a model where you could self-service provision resources that cost about $50 per month and completely replaced a model that was costing us tens of thousands a year. It was an easy win, however... the organisation demanded vendor assessments, compliance paperwork and a billing model which, of course, was favourable to them. But Microsoft's model was "chuck your credit card in and off you go", so that's what one of my colleagues did. And paid for it himself, entirely out of his own pocket in order to save one of the world's largest companies money. My point is that I've done time on the inside and I understand the barriers organisations put in place "because reasons". I touched on this in the June post about the upcoming domain changes:

To be honest, the experience with the public API keys has taught me that it's usually not money that's the barrier to using commercial services, it's corporate procurement bureaucracy. Onboarding documentation. Vendor assessments. Tax forms.

And so too, I have the experience from the outside having regularly received requests to invest hours doing manual labour for the sake of something an organisation is paying a few bucks a month for. That simply doesn't scale and the whole point of providing services like this at volume is that you can go and set everything up yourself with nothing more than a credit card. This one came in while preparing this blog post:

My company is looking to purchase an API key so we can automate user lookups on your site. Our procurement process is wildly complex and I was wondering if we have the option of submitting a Purchase Order instead of using the Stripe credit card payment method?

If this situation resonates, you have my sympathies and my own corporate bureaucracy scars are still raw! If there's more we can do to ease the onboarding path without creating manual labour on a per-customer basis then please let me know. I'm sure there are improvements that can be made, the last thing I want to see is you ending up like my old mate from Pfizer 😞

We've tried to do everything possible to remove barriers. We've made significant investments in legal counsel to get the terms of use and privacy policy right and we've tried to provide answers to all the regular questions in the FAQs. We've even publicly provided a W-8BEN-E US tax form which was often requested by folks in the US. But it won't be enough for some organisations, which is why we do exactly the same thing as Pfizer often found themselves doing which is to provide an enterprise-orientated process where we deal with all this rigmarole... and charge accordingly. If that's you, then get in touch with me.

But What About...?

There will be lots of "but what about...?" edge cases. Let me give you some examples and our views on them:

But what about addresses that don't actually exist?
For most data breaches, email addresses are extracted using a regular expression run over the entire corpus of data. You can see what this looks like in the open source email address extractor used to process breaches. So, what is an email address? Per my earlier explanation, it's anything that matches the regex when run across the breach. That could mean strings that aren't actually an address on a domain get caught up and reported incorrectly. It happens, but there's no way to practically stop it and it's extraordinarily rare.

But what about email addresses from years ago that still appear as breached on a domain?
The argument here is that whilst these are genuine addresses that did indeed exist at one point, they aren't really relevant anymore either due to their age or the address no longer existing (e.g. ex staff). I have both a philosophical and a technical view on this, with the former being that data breaches are immutable. At a point in time, addresses were exposed, and that fact can never be reversed. As for the latter point, those addresses remain in a storage construct we need to continue to support, and every single domain query needs to pick those addresses up and return them to the code processing the search (the design of HIBP means that Azure's Table Storage returns the entire partition on each domain query). Further, in most cases, that doesn't change the total number of breached accounts being a reasonable metric for organisation size and subsequently, the pricing tier they should fit into.

But what about old breaches I don't care about any more causing me to require a higher plan?
It's a similar answer to the previous point insofar as the immutability of history and the need to store the data. It also remains the most reliable metric we have to determine the size of the domain and in many cases, the organisation that owns it. Think of this measurement primarily as a means of slicing up the corpus of data within HIBP and distributing the cost as equitably as possible across the organisations using the domain search feature.

But what about people who don't want to use a credit card?
I'll give you a two-part answer on this, beginning with the recognition that cards can pose legitimate challenges for some people. Just as I was drafting this blog post, someone trying to sign up to the public API reached out after failing to subscribe multiple times with different cards:

For a variety of reasons, I believe the guy is legit, but Stripe reports two payments declined by his bank and another due to an invalid CVC. But using Stripe doesn't just mean credit cards, it also means Apple Pay and Google Pay, WeChat Pay in China, EPS in Belgium, Afterpay in Australia and a raft of other payment mechanisms in different parts of the world. It's hard to imagine a legitimate case where someone does not have access to any of the available payment mechanisms, which brings me to the second part:

The reason we don't support the likes of anonymous cryptocurrency and rely solely on fiat money payments is that it very quickly weeds out the bad actors. That was the whole rationale for putting a payment gateway on the public API back in 2019 - to cut out the abuse. It turns out that once you have to pass the sort of KYC barriers financial institutions put in place, people don't misbehave under their own identity. And yes, there's always fraudulent use of cards, but Stripe has gotten so good at handling that (we pay for their Radar service as well), our dispute rate is only one in many thousands of transactions.

But what about [other reasons related to calculations and costs]?
Amongst the corpus of 12.6 billion records, there will be anomalies. It'll almost certainly be sub-1% and the anomalies won't be evenly distributed across domains; they'll affect some more than others. It's infeasible to ever get that down to zero and it's also infeasible to respond to every single request I know will come through asking for an anomaly to be rectified. The most practical way we could find to deal with this is to keep the pricing structure such that anomalies will be unlikely to have much impact of consequence.

We're also conscious that some people will challenge the cost and it happens all the time with the existing public API key either because of the individual's position in life or the nature of the organisation they work in. But this is why we've structured it as we have, with the majority of domains being within that free tier and the entry level cost being the cup of coffee that gets you access to things like API level access and formal support. This was the most reasonable, equitable model we could come up with and I hope that shines through in the explanations above.

Summary

I know there'll be individuals with catch all domains that have ended up in a couple of dozen data breaches and they think paying $3.95 to see them is unreasonable. I know there'll be organisations with much larger numbers who feel it's unreasonable because similarly sized orgs are more profitable. But I also know that I've been running domain searches totally out of my own pocket for almost a decade so whilst I'm sympathetic to anyone who now needs to pay for a service that was previously free, I'm also comfortable that a reasonable and well thought out model has been arrived at.

I'm excited to see what people do with the new API. The email address search one is presently requested millions of times a day and people have built all sorts of amazing things with it, everything from corporate awareness campaigns to tooling to help protect customers from account takeover attacks to integration within the corporate SOC. It's cases like that last one where I think the domain search API will really shine and if you do something awesome with it, please get in touch and let me know.

I know this was a long read, I hope it adequately explains the rationale for the subscription service and that you use it to do amazing things 😊

You can get started right now from the domain search page on HIBP.

Update: Following feedback and consultation with a range of existing users of the service, we now provide a model for the education and non-profit sectors. See the KB titled Do you provide discounts based on the nature of the organisation? for more information.