Tl;dv: Over 180k meetings left wide open | Hacker News Reader

Tl;dv: Over 180k meetings left wide open

bobdahacker.com

633 pointscolesantiago195 comments

Seems like they fixed this a few days ago: https://tldv.io/blog/our-thoughts-on-the-darkreading-com-art...

But they try to play it off as though this were public data:

> Public sharing settings across AI and SaaS products have surfaced similar findings in recent months. Anthropic addressed exposed public artifacts across Claude and its MCP ecosystem via Google Search.

Also, interesting, they are SOC2 compliant [1], proving again that SOC2 is meaningless/useless.

[1] https://tldv.io/features/security-commitment/

Is there an entity that can validate they are not SOC2 compliant outside of their claim?

Yes, SOC2 require an audit by an independent auditor, and in principle you can request their audit report from them.

Just email the CTO about it ;)

On a personal note, I recognize that I should have kept the researcher updated after his initial outreach earlier this year, and I take full responsibility for that communication gap.

They make it sound like it was a single email. What about all the other outreaches the researcher made to the CEO over a six month period?

Interesting how the CEO didn't contribute any explanation to the blog post and left the CTO out to dry.

Once again proof that SOC2 is nothing but a marketing tactic, and busywork

VC runway afterburners, engage!!!

I used a product with SOC2 certification, which uploads all your chatbot conversations to a server they control (mandatory), potentially including source code and other proprietary data, which can be made visible to public with a single click. Doesn't matter if you are an individual or enterprise user.

They do have enterprise level controls that let admins turn this off. Unfortunately, it is on by default, and some of those basic security controls require a higher tier of service.

It is absolutely wild that these companies treat security like an afterthought. And I also realized SOC2 Compliant meant absolutely nothing.

SOC2 requires a company to write policies in a large number of areas, and to demonstrate that they're complying with the policies they wrote. AFAIK SOC2 does not require anything meaningful about the actual contents of the policies, nor does it require the policies to remain constant.

This is true. They (the auditors) usually have guidance or areas they need to see policy address, but not specific implementation details.

Using common, overbearing MDM or endpoint tools make providing evidence of adherence to policies easier, however.

Reminds me of ISO certification. Where all it does is that your complaints are called non-conformance.

I used to work for a company seeking SOC2 compliance. They told me that I had to install corporate malware because of the compliance. I didn't want to install it on my personal computer, which I had been using for work. They sent me a company computer. I installed the corporate malware on that one. I set the company computer aside and continued working on my personal computer. No SOC2 compliance was harmed in the process.

The company computer typically comes with some data protection agreement, where you agree to only access confidential data via the company computer, or at least to never copy confidential company data to personal devices.

On the one side you are liable because you are accessing company data on your personal computer. On the other you are liable because the company is forcing you to use insecure devices to access company data because of compliance. I think we need to check the contracts to see which liability to choose that will give you the least amount of headache.

You're not liable for using company devices to access company data, regardless of if it's insecure or not. Don't use personal devices for work

I know, I was being coy, but if I'm being honest using company devices makes me really anxious, it's always in the back of my head that someone is going to abuse that outdated and opaque VPN stack from Fortinet or that Kaspersky Daemon I needed to install using a script some guy from security sent me with a Google Docs link over DM, and I'll have a really hard time explaining there was no mishandling of data from my part

Besides the downplaying and obfuscation about the timeline on the first half, I find the inclusion of anthropic and zoom examples to be wild. Just spraying in all directions.

I saw an YouTuber the other day sharing their "day in the life" as an Amazon Software Engineer while promoting (as part of a paid sponsorship) the AI note taking feature of SoundCore headphones, claiming they now record their meetings and receive an AI summary at the end.

I wonder how many companies realise these devices that appear as "headsets" are now funnelling their meetings into these new AI companies who are more worried about the World Cup then replying to security researchers.

Not trying to be rude or mean here but it should be, "a YouTuber" since YouTuber starts with a consonant sound.

You are correct but it is not based on the sound, just if it starts with a vowel. If one listens to people speak, a common mistake, even I do it from copying the same in others, (depending on what crowd I've been around a lot,) is using an before hour as in 'an hour' though most times people will correctly write 'a hour'.

Who writes "a hour"?

Merriam-Webster seems to specifically disagree with you on both points: the usage of "a" vs "an" is based on the sound of the following word, and "hour" is preceded by "an" since it sounds like a vowel despite starting with a consonant.

> You should use a when the word that follows sounds like it begins with a consonant (‘a dog,’ ‘a balloon’). You should use an if the word that follows sounds like it begins with a vowel (‘an ant,’ ‘an elevator’). Remember that sometimes a word will begin with a vowel but it sounds like a consonant (‘a one-time deal’); in this case you should still use a, rather than an. And in some cases a word will start with a consonant but it sounds like a vowel (’an hour’); in this case you should still use an, rather than a. [1]

[1]: https://www.merriam-webster.com/grammar/is-it-a-or-an

Of course there is the other can of worms of debating whether Y is a vowel or not. Also, “YouTube” sounds like it starts with “u” sound, despite starting with “y”.

Maybe English is just broken. Words starting with a long “u” sound make me question the rule - different than other long vowels…

Does one play “a ukulele” or “an ukulele”?

We need an umpire to make the call (;->

There isn’t really any debate about “Y” - it acts as a vowel in most cases, and as a consonant in some cases, particularly at the beginning of a word, as in “Youtube”. The “you” sound is a consonant sound, and that also answers the ukulele question: it’s “a ukelele”, because it’s the sound that matters here.

Using a more familiar example might help illustrate this: we say “a user”, not “a user”; similarly, it’s “a European”, not “an European”.

Here’s a good article about it: https://phonicshero.com/when-is-a-consonant-a-vowel-and-why/

That's interesting only for the fact I recall it being taught different or perhaps missed that all through school. Being based on sound brings yet another aspect of english into the nightmarish realm. My mind races to the many words around my locale and various circles of people I worked around where the lead vowel is silent / near silent ... words like articulate and yet aware that some do go to the effort to pronounce with a high emphasis on the lead vowel like they are were educated / lived elsewhere.

he recorded his comment with an AI service :shrug:

I don't really understand why these "AI notetaking" services/features still exist outside video conference software today -- any company that needs it would have already purchased the license from Zoom/Teams/Slack which takes care of everything including storage and compliance.

I remember while at Amazon, during the "everyone needs to be using AI in every way" push, one of my teammates was proud to present that he had figured out how to split the audio on his mac to record the meeting (because our meeting software didn't allow for recordings without consent) and fed this into speech-to-text and generated summaries of the meetings without any of us knowing (i.e. no consent).

I let him and our manager know that he'd just violated eavesdropping laws across state lines (he's in D.C. and I'm in WA; RCW 9.73.030).

It's amazing how many people don't think that "AI Notetaker" is the same as "I secretly recorded this (as possibly violated the law)"

DC is a one party consent "state", so I don't think Washington state's law would apply to him

I'm very torn on these laws. On one hand I understand the purpose, but on the other recoding phone calls and meetings is immensely useful. On Android I had an option to automatically record phone calls and it has saved my ass many times, while on IOS it's not possible at all.

In many places it's perfectly legal to record interactions you are a part of without informing anyone about it. Publishing them is usually another matter.

Btw, how do you understand the purpose of this law? Because I can't imagine it serving anyone other than abusers and liars that want to abuse a lie without leaving evidence.

Two party consent states are important IMO because if one person records, one person controls the facts. They could release a snippet of the call. But if you get an audible notification that the call is being recorded, you might record it too and have the full context. Just my 2c

If the recording is cut, nobody in their right mind is going to believe that only this was said. You don't control anything by releasing partial recording. You even reveal a lot about yourself because of what part you decided to keep/show. But what was said, was said, regardless of how small part of the whole it was. And you can't simply claim that you did not say it.

Two party consent just means recordings are mostly useless because it lets the worst party control the availability, so unless they are very stupid, they just won't consent.

Interestingly, Zoom and presumably others are logging and reporting on device SIDs and names, so those AI headsets are in server-side data associated with specific workers. I imagine Zoom would rather not advertise this data collection openly — and I don’t have personal proof that the data is transmitted to their servers, so perhaps it’s just collected at the client and then not harvested, unlikely as that may be. But that worker may find themselves targeted for termination if their employer’s platform discovers their unauthorized use of non-employer AI to process employer private work output by their Bluetooth device name.

Ai notetaking is such as signal for stupidity but I'm not surprised that it is getting so much traction. It is basically like getting all your lecture notes from the A student so you can fuck around instead of paying actual attention to anything, and hoping the A students understanding of the lecture somehow rubs off on you.

Kind of funny how that is the pitch for a lot of these products. AI to help you take notes in meetings. AI to help you pass an interview. AI to read and write for you. AI to let you be a lazy bastard, in other words.

Well, on the plus side, SoundCore (Anker) are at least reputable enough (even though security teams would still want to review that!):

Under the security part, they do seem to at least be compliant with your typical security standards (HIPAA, SOC 2 Type 1, etc.) and claim that all data is deleted within 12 hours of transcription. So it isn't like the contents of your meeting are being siphoned off to some fly-by-night service, at least.

https://www.soundcore.com/soundcore-work-ai-voice-recorder

The one thing I know Anker for is that they were uploading the feeds of their doorbell cameras without encryption (even though they claimed e2ee), and iirc they were also uploading thumbnails of these cameras to a publicly accessible S3 bucket, and consistently lying about it all.

So no, I don't think they're very reputable at all.

I moreso meant the quality of their hardware, though that's fair, I didn't recall that. Still, if they are bothering to go to the effort of getting audited (granted, I don't like reading audit reports so I didn't look at who actually performed them) - it seems like they are actually trying, as opposed to what you might expect from a less scrupulous company, where it's more like the wild west.

This should be the kiss of death for any company. The exposure of sensitive data like that, and for that long? There's a serious disconnect between security best practices and law, and how many companies actually operate.

My own company is a sitting duck for hackers right now. I've begged them to implement basic 2FA for 6 months and all they do is brush concerns under the carpet. No one gives a shit, all the way to the very top.

Don't worry, I'm sure this was all an AI agent's fault, so no one to blame and all they need to do is update their code review prompts to not make mistakes.

Also add the word "secure" a lot.

AI agent: sorry for that, I'll build the next version will be the most secured one

Make no mistakes

Actually they are taking one from Anthropic's playbook and saying it's the user's fault for misunderstanding what "sharing" means.

"Government meetings from 23 countries: Brazil, Colombia, Peru, Ukraine, El Salvador, the Philippines, Chile, Indonesia, Mexico, the United States, Qatar, Malaysia, Uzbekistan, Sri Lanka, Haiti, South Africa, Jamaica, Honduras, Argentina, Thailand, Japan, Israel, and Belize. "

oof

Hegseth loves this app!

> He responded within minutes: "thank you! can you report it to our CTO and we will look at it immediately?"

Why could he not speak to HIS ceo himself instead of asking Bob to

I saw that and chuckled.

> Why could he not speak to HIS ceo himself instead of asking Bob to

You can't expect a founder/CEO to spend 2 minutes relaying an email with critical security information to his own CTO, he's surely way too busy disrupting and pivoting and doubling down on product market fit.

I've know executives like this; at some level they seem to be self aware enough* to realize that any message that passes through them will be garbled beyond recognition and so actively encourage people to route around them.

* I know they're just saying the words a self aware person would say to give that impression, but it can be eerily convincing.

This is bad. I run a company in this space (deepfake voice phishing), and one of the most common pushbacks we hear from buyers is: “Where are attackers going to get audio clips of our employees?” ... excluding senior leadership, which most companies already recognize as a risk.

Another similar incident that happened recently was 4TB/40,000 contractors voice + government ID + selfie leaked .. https://oravys.com/blog/mercor-breach-2026

PS: To demonstrate how this can be exploited with real time voice changers i.e. a voice phishing simulator .. we also built a free tool that shows this attack combined with someones voice ..

https://www.callstrike.ai/voice-phishing-simulator (Voice Phishing Simulator)

https://www.callstrike.ai/deepfake-security-training (Deepfake Video Simulator)

It’s obviously a heavily restricted PoC, but it helps demonstrate the attack path in practice.

> 4TB/40,000 contractors voice + government ID + selfie leaked

Leaked selfies? Do you mean ID photos?

Selfies are used during live ID verification. (All of this is supposed to be encrypted, and destroyed within certain regulatory bounds.)

Genuinely thought you meant you carry out deepfake voice phishing.

Our customers do, against their employees using our platform.

I keep being amazed how most basic things are not checked. Cross-tenant isolation is one of the main things I check for... With other generic information leaks.

Sturgeon's Law is proved correct time and again. Most things are crap. Most people produce some crap in their lives. Some people only produce crap. Those people still need to eat but unfortunately some of them (somehow) find their way into tech and actually convince people to pay money for crap.

Especially with a low bar to entry like what is essentially AI-backed transcription-as-a-service, I'm not sure 90% is high enough. There will be 100 companies offering essentially the same thing and it's unfortunately the responsibility of the customer to find the one written by someone who doesn't have a parsnip where their brain should be.

Fortunately we have LLM's to not produce that crap... wait, those LLM's were trained on the existing crap and produce the same crap... oh no.

I doubt SOTA models nowadays are going to produce an implementation without any kind of authentication like here, and not tell you about it.

And even if, a later "is this ready for release" will probably surface such obvious issues.

I do not think LLMs are the problem here. Today, they are most likely more competent than whoever set this up.

Still the six month wait time when everything should be good? /s

No, what I said is that I think this has already been good for some time.

However if you start current SOTA models out on a bad codebase they will happily write more bad code to fit in with the "conventions" of the existing code. Including authentication and isolation. If you start out your app on the wrong foot (for example because you lack the vocabulary to express what you need) you can end up with nicely polished turds

Asking the LLM for a review of the code would still have caught it

I love the never-ending "SOTA" treadmill used to defend anything and everything these LLM tools shit up. Doesn't matter what happens, it wasn't one of the "SOTA" models (which changes every 7 seconds) ergo it's irrelevant, how very convenient for the AI pushers!

Can someone give me exact time when SOTA stop being good? Is it a day, week or a month? As such can I consider anything older than that time period to automatically be crap?

More than half a year ago, someone may or may not have used AI to implement this insecure backend.

That is not a very good basis to start complaining about AI producing insecure code.

If you still want to do it, at least check if it actually is the case with reasonably intelligent current models.

They were RL trained on verifiable rewards. It's not purely learning to predict the next token of a human produced stream.

This is not an issue. IF you prompt it right you could have avoided this with models at least a year old.

I'm very intrigued by AI note takers, but I'm absolutely unwilling to expose me or my clients to this exact problem.

The solution (theoretically) is a purely local note taker, but I haven't found one that's any good. Tried meetily and others in the same vein, including briefly rolling my own. The breakdown in the pipeline seems to be reliable local diarization and speaker identification; even if the transcription is good, when speakers aren't accurately identified and speech isn't well grouped, there's no rescuing it in the summary step.

Drafts.app is hideous but it has great routing capability and a dictation feature. I use it to capture what my thoughts and route based on content. I have a button that routes to an internal voice agent named KiKo. Ideas get routed to Things or todoist. Issues get routed to github, etc. It’s one universal surface for note capture.

But man is it ugly.

My ideal use case is to pipe audio from both my microphone and capture system audio so things like our weekly team standup or my 1:1s with my devs can all have reliable, decent notes without taking me out of the flow of the conversation.

I think clearly the _leader_ in the space is granola, but I'm just not going to use a cloud provider for this.

Drafts have anything like that?

I'm definitely biased as the developer, but maybe try https://whistle-enterprise.com and see how it works for you.

It's a hard problem I've been working away on for a while now. It's far from perfect but every step brings it a bit closer.

Literally starting my monday weekly standup now, I'll run it and see what's up. Thanks!

Thank you! Feel free to drop me and email (info in bio or on the website) if you run into any issues or have any suggestions.

Seems interesting. What would you say are the biggest missing points currently or things you want to get working/improved but couldn't yet?

Thank you, great question! Hard one to answer, thought about it a lot and it's going to be the diarisation of more than 5+ speakers per audio stream (your microphone + system audio for a max of 10). I actually spent a lot of time that went completely nowhere trying to fine tune my own diarisation model, it was fun to a degree but painful to see my output end up worse than what I currently had after days of work. Having a bot join the call would be such an easy way of diarising, the call software has already done it for you, but feels like a bit of a cop out.

Two more improvements, audio quality improvements which is currently in the works and close to release and a new document generation model. I'm currently using a custom fine tuned Phi-4 (released December 2024!) model, that's _so old_ in the grand scheme of LLMs, I just haven't had time to benchmark and properly test some new models whilst this currently does a good job as it is. There has to be some gains here, but who knows!

Interested also, which text-to-speech model do you use? For diarisation Granola uses a chrome extension instead of a bot if that can give you ideas.

I think this is just to capture the audio to ship off to their servers and do the crunching. You mentioned the word "extension" and thank you so much, I've been thinking about how to do integrations as it's something a few users have mentioned, but keep the whole "completely offline" angle. I could build standalone extensions that integrate with it if it's something a user wants. So damn obvious in hindsight!

Ahh yea as for the models:

Speech to text - Nvidia Parakeet TDT 0.6b V3

Diarisation - Nvidia Marblenet for the speech detection, TitaNet-Large for the embeddings and then using NeMo multi-scale to do clustering around them

Since the participants are known and limited, have you tried building around samples tagged with user/person names?

I tried something similar by putting together a "global ledger" of speaker identities. This was more happening during the call than having a predefined one but I just couldn't get it to work properly. The issue being that as soon as some speech gets assigned to a new label or an incorrect one, everything ends up getting misaligned and gets messy quickly.

I might take another look into doing it in a different way that gradually builds up from successful calls, I just need to think of how to do this in a simple(ish) way for non-technical users and a way that still works well enough on low to mid tier laptops.

I know this is hn and not a product development meeting, but:

If you’re targeting data for people to use on the same call then requires more intense work, while if you’re targeting data for people to use after or on an ongoing basis (eg an established business meeting with staff/vendors/etc) then having a predefined one seems good. Same interface as a CRM: picture, some audio clips for users to choose from, and confidence scores on each section of the recording for cases where people sound similar or are talking over each other. Over time the diarization gets better as users accept/reject/tag samples and that work helps them feel more aligned with the tool.

100% looking at it from an after use basis, whilst there is live transcription, trying to do anything above that on 6 year old laptop CPU just grinds stuff to a halt.

Some great suggestions, appreciate you taking the time to write it out, it's given me some things to think about.

> I'm very intrigued by AI note takers, but I'm absolutely unwilling to expose me or my clients to this exact problem

Unfortunately it’s mostly not up to you. It’s a weakest-link problem. It doesn’t matter if you don’t use a note taker AI, if even one person on the call uses one. Their tool doesn’t notify you and usually the person doesn’t either.

It also has the reverse impact to the person using the note taker, where people say less around them. Same as if I'm talking to someone with Meta glasses.

I wonder if the people who use these tools know the people they meet with speak less during their meetings, and then all of the participants have a post-meeting call without them to say what they really thought.

> It doesn’t matter if you don’t use a note taker AI, if even one person on the call uses one

Yeah, but I'm unwilling to be that person.

> The breakdown in the pipeline seems to be reliable local diarization

Yep, diarization just hasn't been well solved yet. As soon as it has, the quality in note-takers, meeting transcripts, etc, will sky-rocket across the board.

That’s a very fair concern. I completely understand why you’d want to avoid exposing yourself or your clients to that risk.

I also agree that diarization and speaker identification are probably one of the hardest parts to get right if the speakers aren’t separated correctly, even a great summary won’t fix it.

We’re looking into more privacy preserving approaches for MindNote (a multimodal AI Notetaker), including local/on-device processing, so this is really useful feedback. Thanks for sharing your experience!

My biggest worry is the small talk that you casually have in meetings. Comments about your pattern of life, family, locations, friends and health. Slurping all that up over 100s of meetings is concerning. Stored raw transcripts of meetings is a huge liability. Should redact small talk and only store useful extracts.

Some notetakers filter that stuff out automatically.

Six Months !?! If I'd left a vulnerability like that open for 6 hours there'd be hell to pay. Something that critical is call for hitting the big red off button.

In this case the CEO was aware of it and... did nothing.

Ah, see, if the CEO does something that hurts the company it's ok. In fact, they will pay him (or her) for making the mistake in the form of a golden parachute. It's only the rank and file employees that actually suffer consequences from failures.

Not the first time I read a shitty implementation with Firebase, I'm not blaming the platform, but seems there is a huge skill issues around it.

Wasn't a dating app exposed this year with same negligence or firebase security?

It's almost like people need knowledge and experience to work with tools securely. The problem with Firebase (I think) is that its marketing is "it's easy to use" and I'm confident most problems - like storing this info - is easy to figure out and finish, then move on to the next thing.

But this is lazy / "move fast" software engineering. They mention all of these certifications, I think they should be stripped of them for a year because of a failure to respond / act.

If something happens again and again, it is by choice. Firebase chooses to make it “easy” to get started rather than “secure by default”.

These companies take anything but responsibility. The fact that they’re trying to downplay it in their own blog post with “but look at the others, they’ve had security issues too,” as if that makes it any better, says it all. I would never trust that company with anything ever again.

Funny enough, there is a news story in which the same CEO that didn't care about the breach for 6 months told USA Today 'that European tech needs to be “bold but secure” to win the enterprise AI race'.

https://archive.is/QQy2b

Heh, interesting. I literally just vibe-coded an app tonight that takes a meeting recording, runs it through whisper to generate the text, then through your local codex gives you a summary and creates actions that are pushable to either github (technical) or PM tool (project level)

Takes about 7 minutes for a 2 hour meeting on my 3080 GPU so well within useful timeframe.

I did it because i didn't want to pay £7 a month for a discord meeting notes taker, but seems generally useful. And ofc you could swap out for a local model if you have more compute than i do...

Why not just pay attention in the meetings you are paid to attend?

I just did this - i took notes by hand during a meeting, but also took the AI summary and compared the two. I did miss one item. My hand notes were far better though.

Depends on your level of interaction in the meeting. I have a lot of meetings where I’m having in-depth “fact finding” interviews with one or two other participants. Taking notes breaks the flow and sometimes the other participants find it rude. So instead record those meetings and generate a local transcript after using whisper.

OK ill rise.

It's not about the meeting, its about the work after the meeting.

This tool creates a list of reviewable actions, and has API access to github and project management tools so i can tag the owners and submit them direct.

It also creates comprehensive, well versed minutes, in a historically logged list.

How many of your meetings you pay attention to do you actually get all that output, that you can share and push back to your team?

Hmm, does Ukraine know that Russia is watching the Ministry of Digital Transformation's meetings?

The worst part of these AI meeting notes and recordings is that I have literally never seen anybody, in any situation, go back to them. The (very) few times I ever bothered to check the transcripts and especially the summaries immediately after a meeting, without fail they'd have something in them that is the complete opposite of what someone said in the meeting, or they'd miss extremely important clarifications or context. Literally worse than useless, actively detrimental, but you bet your ass whichever moron forced these to be on for every meeting is putting it on their resume - "Increased long-term organizational cohesiveness via comprehensive automated meeting notes" or whatever type of bullshittery.

That is a wild claim. Every single quarterly meeting my company has with customers begins with redistributing the notes from the previous meeting, and "to-do" points are regularly referenced.

AI notes are a godsend, because instead of forcing one person to keep careful notes throughout, the meeting's host can quickly review, edit, and send out the minutes - a job that otherwise would literally take more time than the meeting, because one person spent the meeting just taking notes. The AI notes require careful review, and contain a lot of repetitive fluff, but all notes require review before distribution.

So, now you are aware that this is a normal thing in other companies.

> That is a wild claim.

Perhaps we can all agree that usage is very bimodal, between the people who refer to auto-notes constantly vs never?

My own anecdata, fwiw. I have periods of my year where I spend the majority of my time (like 25-30 hours a week) holding and recording calls, which I then transcribe and review. I’ve never once found an error of this type. I also record and review a lot of quarterly earnings calls, and have never experienced an error of this type.

This breach will help with tl;dv's awareness. More people will suddenly become aware of it. The downside is that less private conversations will be hooked up to tl;dv, and more public seminars will be fed instead. This is a win-win overall. And it is a PSA to all other companies to secure their servers a little bit better.

1785854861 | Tl;Dv (Too Lazy; Didn't Validate): 181,874 Meetings Left Wide Open | https://bobdahacker.com/blog/tldv-hack | https://news.ycombinator.com/item?id=49169753

1785962536 | Tl;Dv (Too Lazy; Didn't Validate): 181,874 Meetings Left Wide Open | https://bobdahacker.com/blog/tldv-hack | https://news.ycombinator.com/item?id=49188723

I think this is one of the few times public disclosure wasn’t a good idea. Some of these are government meetings and could put lives in danger.

Also, shame on the CEO for not making this an emergency and confirming it was fixed by the end of the day.

> tl;dv names their microservices after pasta. A subdomain scan reveals cappellini, carbonara, fusilli, pasta, penne, puttanesca-v0, and ravioli, all under tldv.io. An entire Italian restaurant worth of Express servers.

Pretty appropriate, given a vulnerability of this severity. Literal microservice spaghetti.

(Also, please, let's all move back to boring names for services and servers. Nobody likes trying to decode what all these silly names mean.)

EDIT:

omg, the disclosure communication is infuriating.

> We're on it. It needs some time, but rest assured we're following through. For further communication, i'll recommend reaching out to our CTO

This should have been a P1 that was fixed same day, and they strung him along for months. Absolute amateurs.

From an online article about the CEO:

"Raphael Allstadt, co-founder of Germany’s fastest-growing startup, tl;dv, argues that European tech needs to be “bold but secure” to win the enterprise AI race."

"But Silicon Valley may have more engineering talent on paper; Europeans care far more about data, security, and privacy. That means our builders pay closer attention, build compliance in from the start, and are ultimately better suited to serve the enterprise market, especially here in Europe.”

As tl;dv scales, Allstadt’s mission remains twofold: to prove that a European startup can out-execute the US giants on product, while maintaining the privacy standards Europe demands.

The irony

Funny, my first thought was that this has to be related to Firebase...

Turns out building https://github.com/sleep3r/crispy was a pretty good idea

I'm starting to believe Firebase is cursed at this point.

We're going to open your database by default, but PLEASE remember to secure it later! Footgun deployed successfully.

at this point, it's a feature, not a bug.

> I wanted to say something so badly. "Hey, you might want server-side validation too." But this was a proof of concept, not a consultation.

oh man, imagine telling them "you don't want people like me breaking into your app! Just look at how I got into this call!"

The worst thing is that there is nothing technical preventing this kind of service from running fully local. All the audio is already being routed to any participant and transcription models are lightweight enough and can run in virtually any computer.

I love the fact that, if the emails listed in the post are correct, bob didn't even try to be professional with formatting or capitalization or anything. That shows how much power this man had in this situation lol

We use a similar tool at work and it's automatically invited to every meeting. It records, transcribes, and lets you search for something somebody said and when. It is incredibly unsettling.

My prediction is that soon every mid-size+ company will have their own hardware running an open-weight model and all employees will be required to connect to it and it alone for inference.

OK, the issue sucks. That said, the post was written by an LLM and It's not pleasant to read. If I didn’t work with Claude every day, I might feel differently, but because I do, this reads like slop.

TBH I've been finding Claude quite funny lately, the pasta jokes are pretty good.

> The irony is al dente.

doesnt change the fact that the vulnerability existed. Stop whining

The disclosure section is gold. It’s such an accurate description.

Major consumer companies reply just like that.

It’s incompetence and I think to an extent also arrogance.

My first thought was "huh, I wonder if they follow GDPR, that starts sound to like Article 32 violation" (related to security of processing). I checked the privacy policy and yes, they are based in Germany so GDPR likely applies to all of their processing.

However that privacy policy raises also quite a few concerns. They say that "profile image URL" is based on contractual obligation which is quite weird, just why profile image is necessary to fulfill a contract? Additionally IP address and location are collected for "adapted pricing" and it's "legal and contractual obligation". I can somewhat understand that if it's used to calculate VAT, but "adapted pricing" sounds much wider thing. And even VAT calculation itself isn't really "adapted pricing", it's something that the company needs to handle. They are of course free to change the price based on that, but that price change goes more to legitimate interest rather than legal (or contractual) obligation.

They also claim that "Product analysis and improvement, marketing and attribution, incident management and in some of our logs" as well as "Analysis of products and navigation on the site and application" are also "legal and contractual obligation". I honestly want to hear what contract necessitates those or exactly what law requires them to do that.

Is this still active? I wouldn't mind spying on some meeting notes. Sounds fun.

"Lets circle back and touch base to tease out any low hanging synergies we can capitalise on" - repeated 181,000 times

"You're on mute"

"Nothing on my end"

If they haven't fixed it in 6 months, I'd say it's fair game to scrape as much as you can.

I know this is a joke but it's still a felony, for your own sake don't do this.

Serious question: how is it a felony?

Distributing it might count as copyright piracy, but merely downloading public data?

Just because company failed to limit access does not mean it is public data. Just like stuff not being screwed or glued down doesn't mean you are free to take it.

The federal government outlaws "knowingly accessing a computer without authorization or exceeding authorized access" to obtain information from any "protected computer" (which, in this context, means any computer involved in interstate commerce - which, in practice, has been ruled such that it certainly includes any Internet-connected server of a corporation).

Your defense would have to be that you were authorized to access the data, or that you did not know that you weren't authorized to access the data. Not merely that the data was easily accessible.

I suppose you might be able to argue that everyone is authorized to view the data if they made no attempt whatsoever to prevent you from doing so.

It’s not public data. Just because it is unintentionally publicly available does not make it public data, except in the strictest, most pedantic interpretation (that would never hold up in front of a judge).

the prison time might be the best part of reading through 180k meeting transcripts

How is this even a product? Call me cynical and maybe slightly jealous but I increasingly believe the difference between successful and unsuccessful startup businesses is suckering investors out of a check and faking it til you make it. Influencer trash vibes.

holy crap. how do you respond as CEO to this and not escalate to like priority #1?

then kick the can for 6 months?

A post on LinkedIn where this CEO seems very active should solve that.

> how do you respond as CEO to this and not escalate to like priority #1? then kick the can for 6 months?

We might be able to check the meeting minutes and get the answer?

To forget tenant isolation on one endpoint is bad enough but to ignore it for 6 months is madness. I am at a SaaS company and our customers have such strict security requirements for us and that is for less confidential data.

Wow.

This is essentially the #1 Firebase footgun. Having client-accessible databases with optional rule-base security has always seemed a bit dumb to me.

Now let's see if European users get their GDPR article 33 notification of the breach...

"Because the common denominator between both of these distinct incidents was Firebase, we are taking the additional step of immediately removing it from our tech stack altogether to definitively eliminate the risk of similar vulnerabilities in the future." - from their post-mortem.

Thank god the root cause was definitively identified! /s

So did he email privacy@tldv.io? Why not? Maybe someone who understands it would read it.

It's unclear. Only stating the existence of the privacy email.

> [...] Buried at the bottom, a single line: "If you have discovered a privacy or security issue that we should address, please always let us know at privacy@tldv.io. Our security team will respond within 24 hours." I emailed the CTO directly. Six months. No response. [...]

This is near the disclosure schedule

From the article - he reached out to the CEO directly, who acknowledged and said it was being worked on by the CTO. He did this repeatedly over 6 months.

I wonder how the HN algo chose this one link to Bobs blogpost to be the one to hit the front page. This story was committed three more times by three different users, 6 days ago (16 points), 4 days ago (11 points) and another one 4 days ago, again with 16 points.

I understand the need to shame this platform, but why expose all their clients to this much risk? This disclosure here just named a whole bunch of clients. Why?

You need to read the article.

I read the entire article. Did you? There’s no reason in there to expose this company’s clients.

Edit - Are you capable of answering my actual question or was that the best you could do?

He has been emailing the CEO for six months with no replies. This is has also been posted here before with not a single pip or comment ... :-)

And these customers absolute lack of technical due diligence, on this nth example, of move fast and break things...makes them deserve what they are getting.

> And these customers absolute lack of technical due diligence, on this nth example, of move fast and break things...makes them deserve what they are getting.

That’s a garbage take. These customers didn’t move fast or break things. They trusted a company that made a promise and that company let them down.

>> They trusted a company that made a promise and that company let them down.

That is a Boeing and Volkswagen type of excuse. The engineers did it!

No it’s not. Blaming customers is Volkswagen saying it’s your fault for buying a TDI.

I totally trusted this vendor selling me this snake oil, and he promised me it would work. I can't believe he let me down. This is all his fault.

Security incompetence doesn’t turn basic meeting recording and transcribing into snake oil.

Technical due diligence do not including pep test!

Read the article again, then. Anyone that has could get the list with a trivial amount of work. Security through obscurity isn’t going to hide that client list.

And who knows? Maybe someone competent whose company is a client will see that list and say, “hey, boss, I was on HN today, and…”

I think it is fine, the person hasn't really leaked any critical information. He named a few clients that are mostly government departments, and it would be a public interest concern if said issue is ignored for 6 months anyway

As I see it he is not the one exposing clients to risk. He is frustrated that no one is fixing it. The company that left themselves open like this are the ones that are exposing their clients.

If this person is doing his best to do the right thing, there are probably other people who know about this vulnerability and are using it without telling anyone.

What's the alternative? Seriously. He's spent six months trying to get them to fix it. The risk is already there.

Right. At a certain point, the customers and investors and world need to understand the scope of the negligence so they can prepare for their own fallout of having that information fall into the wrong hands. Public disclosure is a responsibility of being a good citizen and engineer.

I’m so sorry for tl;dv ‘s insurance company.

Sometimes shame is only option to get things fixed. Sadly we do not have any reliable government institutions that could mandate immediate shut down of services. Before that only way to get things fixed is public shame.

Reply on news.ycombinator.com