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.
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.
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.
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
"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. "
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.
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.
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 ..
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?
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.
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.
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…”
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.
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. [...]
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.
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.
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.
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.
Asking the LLM for a review of the code would still have caught it
oof
Wasn't a dating app exposed this year with same negligence or firebase security?
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.
Leaked selfies? Do you mean ID photos?
then kick the can for 6 months?
We might be able to check the meeting minutes and get the answer?
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.
Edit - Are you capable of answering my actual question or was that the best you could do?
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…”
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.
That is a Boeing and Volkswagen type of excuse. The engineers did!
> [...] 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