Hacker News

z.ai made a statement, screenshotted in this article: https://finance.sina.com.cn/tech/roll/2026-09-18/doc-inisfye...

claude translation:

Dear ZCode users,

We take today's community discussion very seriously. We carried out an internal review right away, and we first want to apologize to the affected users. Here is an explanation of what happened:

The issue stems from ZCode's "codebase indexing" feature. This feature is meant to help users generate a repository index locally, which supports session checkpoint restoration (including past versions), rolling back to past versions, and Repo Wiki, among other things.

When the Repo Wiki feature generates Wiki pages, it may trigger an upload of repository data. After the Wiki pages are generated in the cloud, the uploaded data is destroyed immediately and is not stored. Because this feature was enabled by default in its early launch period, some users were affected. We sincerely apologize for this. The issue has now been fixed.

We understand that any data-related issue directly affects users' trust in a product. We will open-source the ZCode codebase in the near future so we can improve the product within a more open ecosystem. We will also invite third-party evaluators to review how the system operates, and we'll keep publishing updates on the review, building your trust with full transparency.

We deeply apologize for the trouble this has caused. As compensation, all ZCode users will receive one extra weekly quota reset, which will be issued today.

Thank you again for your attention and oversight.

Is it naive to assume that the agent will try and access anything on your disk, either accidentally or maliciously?

Permissions classifiers in auto mode are just models trying to guess if they're doing the right thing.

Claude Code will tell you that it went around a sandbox because the sandbox blocked it. At which point, you ask yourself the point of the sandbox.

While we're on this, I find it really really weird how windows defender insists on sending my codex work files for analysis all the time (which I block in automatic permissions so it has to ask me in a notification). I don't think i've seen it ask to upload more than one or two things, and it doesn't do it with other AI app I use (eg Claude Code) but they really want to see what's inside my codex files.

It's easy to trigger, I just need to go inside Codex settings and change something, it saves and instantly windows defender who never wants anything want to "you may be at risk, let me upload that for analysis yes/no".

Is it much different from Apple and Google who trick user into agreeing and upload all user's data into a US cloud for convenient LE access?

Also, as I understood, this is a feature to allow server-side indexing of the project. But of course I wouldn't run this, and I generally wouldn't run any IDE or AI tools without a sandbox.

Sadly this plague of silent auto-updates is spreading to Linux. For example, browser plugins in Firefox on Linux can silently auto-update without user consent and without any checks and can be used as backdoors. Furthermore, the auto-updates are not using a package manager; firmware also seem to quietly update and also is not using a package manager.

They learned nothing from the Grok Code saga.

If anything, that should have been a learning lesson to NOT trust harnesses, especially new ones.

Tangential, mildly amusing thing I noticed while implementing my own harness: GLM and particularly Deepseek are both fond of trying to read dotfiles and anything listed in your .gitignore files. I only noticed it because I have separate read scopes for project files, ignored files, dotfiles and external files, so the latter three always prompt me for approval.

I'm sure there's a perfectly reasonable explanation for it, which has nothing at all to do with exfiltration of secrets, but it does amuse me when it happens. I imagine the labs have access to lots of secrets that various actors would like to get their hands on...

(shameless plug for my own harness, which is open source and doesn't have a backend to send any data to: https://www.opairdev.org/ )

There had to be a catch to the "free" promotion they're offering this month if you use ZCode. Glad my instinct to isolate it helped me, but I feel sorry for anyone whose secrets, etc. got vacuumed up by Ziphu

Things like that - and other examples posted here - are why I 'm sticking with OpenCode despite it having some papercuts that annoy me.

The incentives are not there for them to do shady stuff like vacuum your files, inflate your token count just because or many other things.

Never use a Harness if it is not opensourced.

DeepSeek Harness is my favorite for coding. Hermes is my favourite for Other things , followed by OpenCode (sucks at managing long running services) .

Others swear by Pi.dev

That’s explains the 300 million of tokens on the weekend only if you use their tool.

Evidently there's not a single inference provider that can be trusted. This is why I don't use OpenRouter either. How am I supposed to trust all of those random providers I've never heard of, when I can't even trust the ones I have heard of? Day by day, the entire industry is hellbent on proving that open-weights and self-hosting is the only safe path forward for us all.

Closed source agents are a red flag no matter if its China or America. Always use an open harness with a good reputation and enough users that someone will notice if they push malicious code like this one here. Right now that's Opencode and Pi.

I've been using ZCode since it's initial release and can't find any of this in my data. There aren't any logs showing capture or upload, and I don't even have a ~/.zcode/v2/checkpoints/ directory.

So unless they've cleared it all with a recent update then it doesn't seem to affect everyone.

Seems like a repeat of the Grok CLI fiasco:

https://news.ycombinator.com/item?id=48892468

https://x.com/a_green_being/status/2076598897779020159

Envelope encryption with server-held private keys turns local backup into remote asset extraction.

At this point does any of us/you really think all those piss-cheap tokens are coming out of thin air? That unlimited token-usage during certain hours was not coming from Chinese side of Himalayan glaciers, was it?

Besides why would you use a closed source harness from a certain place, even if you decide to use the model (if nothing then for the price alone). And, that first remark wasn't just for ZAI but all the providers.

At this point: wrapping the harness around something like sandbox-exec or agent-safehouse is a must. Better still, create a new user account (after so much resistance I am warming up to the idea).

Will ZAI see a blowback after this news? Naah. People will keep using it. Hell, I will keep using it. That's how it is now - post truth and post LLM world.

PS. Anyone singing praise of OpenCode here, it's literally one of the worst harneses, open or not. Just look at their fricking issues - the strategic and rampant placements of "no planned" is mind boggling. And for what? Slightly better than ClaudeCode in token consumption and that too starts getting muddled after a while.

Lots of modern software plays it loose with privacy, but this IMO crossing a second line: doing so with zero notification whatsoever, in a massively intrusive way, against data that is almost certainly private and possibly illegal to exfiltrate, with no obvious way to turn it off.

That crosses into outright malware.

Makes me not want to use GLM or other Z.ai models either, since who knows what interesting easter eggs are embedded in their training data.

You know... (puts on foil hat)... I did notice that Z is also the weird Russian logo for their invasion of Ukraine and Russia and China have cooperated to some degree (or at least China is helping Russia in exchange for access to resources). I dismissed this when I first thought of it, but I will now leave it here. Still probably coincidence but my Bayesian priors were just updated in its direction very slightly.

I recently began playing around with ZCode. Works pretty well. Super sketchy though if it is in fact silently uploading full git history of every user's projects. This is why we need not only open weight models, but open source harnesses as well. Luckily the project I'm trying ZCode on is already open source (Molecule.dev), and I'm already allowing full telemetry with my other agents/harnesses (e.g., Claude) for this particular project, so it's not a huge deal in my case, but it's obviously a huge deal for anything proprietary.

This sounds a lot like the same thing Openai did with navier stokes, but Openai is more stealthy about it.

Built a similar read-scope gate and the fiddly bit was symlinks escaping the project root.

I would never trust these Chinese vendors with their tooling or their own inference endpoints.

afaik DeepSeek also trained on everything that was sent to them via OR and that's why you got that massive discount

This is all publicly available anyways, who cares? Also you're practically consenting to it when you run an agent locally

"Why yes, we had to exfiltrate 100% of your data so we could vectorise it and improve recall by -0.3%"

Oh no, they are going to steal my shit tier slop code their model wrote anyway

Fresh AI slop

The funniest thing is that the uploaded content is encrypted using a key that the users don't have.

Ohhh no another one found that agents don’t actually run locally. We already had the “grok uploads all my stuff to Google cloud bucket” news…

next I can’t wait to see news about “ai company is using my data without my consent” as well.

You know, swival.dev is fully opensource, doesn't hide anything, fully supports GLM, has excellent context management to keep token usage low, and doesn't send anything you didn't ask to the cloud.

Again. You really should stop using closed source harnesses, just because "It's cheap!".

Well - spy agents. Not surprising. But people could have suspected this before surrendering to AI skynet.

Is this a step forward compared to previous distillations or a step backwards?

WTF is token stead this is pure content marketing slop? Genuine question

> However, across the entire policy, FAQs, and changelogs, there is not a single mention of silently packaging and uploading entire workspaces and full Git histories.

So this is criminal hacking, right? It will be prosecuted as criminal hacking? Not in civil court, but criminal court. Because if not... then are we totally done pretending, and we're just openly admitting that computer security law only applies to individuals, and corporations are exempt?

WOW. I actually did buy a month of GLM because GLM-5.3-Flash is so great and ZCode is honestly one of the best harnesses out there from an HCI perspective, and I won't lie, this is pretty gutting. I guess this settles my inner turmoil about open-sourcing my cAI research, at least...

With that personal failing in mind, I'd ask y'all to permit me to toe the guidelines just once, to proffer a hearty nyah nyah told ya so on a comment thread that spawned ~a dozen disagreeing replies this week! More seriously, I think this[1] is highly-relevant, shockingly-underreported context about the extent to which four PRC companies --Z, Alibaba, DeepSeek, and Moonshot-- are acting in bad faith. Consider it testimony as to their character, just in case anyone is thinking this might just be a simple misunderstanding.

So... nyah nyah, told us so:

> In the PRC, they[1] leaked tons of national secrets on the PRC's latest AI campaigns, the inner workings of their "opinion monitoring" (read: performative panopticon) and "stability" (read: violent oppression) departments, Chengdu's whole CCTV network, direct-energy weapons plans, espionage activities in Syria to hunt down Uyghur refugees, and god knows what else that Anthropic didn't divulge to us common folk.

> In the US, it's very clearly an attempt to rip off a competitor. I'm not sure how else you could possibly see it. Even if you're a distillation fan in general (which A. why and B. plz don't), they did this through a network of Japanese and Signaporean shell accounts, presumably at least some of which were abusing Anthropic's subscription service in a ToS double-whammy, as it would be exorbitantly expensive otherwise. They also had to hack around Anthropic's API to get CoT traces, which seems impossible to explain away as anything innocent.

> I've been beating the "China isn't necessarily an enemy, it's gonna take us all to handle AI" drum for literally years, but this attack was just... gross. Gross in scale and gross in arrogance. Not a good sign for the dawning alignment crisis, to say the least :(

> TL;DR: Use these services if you want, but know that you're supporting aggressive escalations and companies that very clearly don't give a flying fuck about violating the law, much less your ToS. So... buyer beware, I guess.

[1]: https://www.anthropic.com/threat-intelligence-report-septemb... is the report.

I lowkey suspect this PRC-based scandal has been underreported because Anthropic went insane with the sidebar UX on this page for some reason; there were many reports on the reports of Houti and Iranian usage, and very few on these sections. Could a week's mass media cycle be this seriously affected by such a stupid thing as a sidebar experiment?? Strange truth, or just fiction?

Crossposting from the other thread...

Tangential, mildly amusing thing I noticed while implementing my own harness: GLM and particularly Deepseek are both fond of trying to read dotfiles and anything listed in your .gitignore files. I only noticed it because I have separate read scopes for project files, ignored files, dotfiles and external files, so the latter three always prompt me for approval.

I'm sure there's a perfectly reasonable explanation for it, which has nothing at all to do with exfiltration of secrets, but it does amuse me when it happens. I imagine the labs have access to lots of secrets that various actors would like to get their hands on...

(shameless plug for my own harness, which is open source and doesn't have a backend to send any data to: https://www.opairdev.org/ )

[flagged]

Elon has nothing to lose on trust.

Z/GLM now has a lot to rebuild.