I like it! Do you have an explainer that’s a bit more readable than the one in the README?

Thanks. A few diagrams on the spatial layout and information flow between stages would indeed go a long way. I'll take that up in the upcoming commits.

Thanks for being so transparent.

Is it understood what the initial attack vector was? How confident are you that compromise is contained to just this GTM application (and lateral movement/exposure hasnt occurred)?

We identified and verified the attack vector and have no indication there is other compromise.

In addition to our in-house security team, we also have hired a 3rd-party firm to do their own investigation and validate our analysis.

We hope to share more about our findings once the investigation concludes.

Fwiw, Anthropic are doing some weird staggered lockup thing, so employees shares start coming online in some strange way much sooner.

I dont remember the details, but its not safe to assume that the first 6m have no employee-shares being sold.

Anthropic's staggered lockup only applies to certain classes of insiders, which does not appear to include regular employees.

It basically allows officers and directors to sell even more of their stock than a normal lockup period would allow for. The only reason for allowing this is if the stock price is expected to drop significantly between the IPO and the expiration of the standard lockup period.

I'm with you- the attack surface of openssh (and my trust in damien) is a better match imo

If you use Tailscale SSH with headscale you are also affected. The vuln is in tailscaled, not the control plane / netmap.

Former tailscalar here and now someone managing tailscale at an org.

Having done my own review of the vulns and the fixes as they are landing in github, its pretty clear to me that this only affects situations where you use ACLs to permit ssh into a node (via tailscale ssh, not via opensshd), but dont give a path to sudoers/root.

If a user had no SSH access and/or connectivity to a vulnerable node, none of these vulns give them access.

However, if you have a vulnerable tailscaled and you give ssh access as some non-root account/privs, then yes, these vulns are relevant as theres now known privilege escalations that give access beyond some specific user account.

Not trying to play down the vuln (its important), but its not a free-for-all RCE that some ppl are acting like it is.

Calling it “tailworm” feels like alarmism to promote their find more.

Would only show worm behavior if machines within the tailnet had the right settings and also were members of other tailnets with those settings but then it would require someone in that tailnet to have another tailnet with the right settings…

It’s far fetched.

But a large portion of Tailscale users should be software engineering professionals, so a lot of the time, people use Tailscale just for SSH. I personally feel the impact is still quite big, especially since SSH leads directly to control of the server.

Do you use tailscale ssh but either dont have sudo, or don't permit login as root?

If all your tailscale-ssh ACLs allow you into an account with sudo, or allow you into root, theres no impact (you cant do anything with these vulns you couldnt do a month ago). Thats the nuance I'm trying to get across (admittedly my english is bad).

On the majority of nodes I provide Tailscale SSH access to, it's as root or a user with passwordless sudo. It's being used for admin access. So yeah, on those it doesn't matter as much.

There are a handful which restrict users to their own user accounts. But it's on dev server with friendly devs, so even for that use case it's not really the end of the world.

I'm sure our infosec folks are going to demand that we upgrade everywhere, however.

Engineers use vanilla SSH for SSH, Tailscale SSH is more focused on enterprise stuff where you can have an audit trail of all commands executed and enforce ACLs.

I’ve used tailscale in quite a few jobs now and not one have we used tailscale ssh. Call us old school or whatever but at my current place of employment we provision ssh keys on a box with ansible and just use regular old ssh over the tailnet, not tailscale ssh

Tailscale SSH is different than doing SSH with Tailscale

The vast majority of tailnets are just hobbyists on the generous free tier and probably don't use tailscale ssh, but rather ssh over a tailnet. Usually it's either just one person or a family sharing access to the tailnet. But for corporate accounts where ACLs are used for SSH yeah this is big.

I'd be fascinated to see numbers on whether your assumptions here are correct in terms of the ratio between hobbyists using Tailscale on their home network and actual paying customers using it for more serious purposes. Does that change if you look at user counts rather than networks?

My gut feel is that you're correct but presumably they do have a decent number of proper corporate customers or none of these features would exist in the first place.

It's just basic economics. This is true for any service that offers a freemium model and you have more than 1,000 total users.

Any time you provide a free tier it's going to have more users than the paid options. I'm sure they make a decent chunk of revenue from the paid users so it's not a comment on earnings/economic feasibility specifically.

My sample size is small but nearly everyone I talked to uses basically zero of the tailscale advanced features. The VPN, the tunneling, and the internal IP/DNS - nothing else. And the same is true for headscale and netbird.

I use tailnet as a vpn and ssh into my server via the tailnet, but I dont use tailscale ssh the product. I think most people fall into this bucket.

Of course it should be taken seriously, but this topic is also piggy backing off of 7 days straight of non stop viral engagement farming on X after php2shell, where people we're farming likes by just posting stuff like

"YoU ThInK PhP2ShELl iS BaD JuST WaIt unTiL TaIlScAlE get HaCkED, ThIS JuSt ThE BeGiNNiNG oF a CybeR ReCKoNiNG"

To farm FUD from normies. Which of course led researchers who were working on tailscale vulns to take advantage of that algorithmic attention as well and talk about their research.

Like Im pretty sure we knew about this vuln for 3 days now, or atleast that it was responsibly disclosed (and not being used in the wild).

The attention this is getting is definately the product of some cringe viral snowball effect of the also rather benign p2pshell vuln.

It feels like there's a ton of AI influencer astroturfing going on, because its wildly easy to get engagement when you add the words "agent/ai" with "vulnerbility" and [insert popular service].

> so a lot of the time, people use Tailscale just for SSH. I personally feel the impact is still quite big, especially since SSH leads directly to control of the server

I too use ssh with Tailscale, but as I understand things, this is only impact "tailscale ssh", which is different from doing "ssh over tailscale" (over tailnet), at least how I understand other comments here, someone please correct me if I'm wrong.

On a tangent: are you really using tailscalar as the name for employees, or is it a typo and you meant tailscaler?

Tailscalar!

Another fun fact: there’s a list of things with tails.txt and scales.txt (see GitHub) and that list is used to generate the random tailnet name :)

(Or at least it used to, it’s been years)

I think this does downplay the vulnerability, since it assumes that teams using tailscale SSH have a lax security posture. I would hope that a security conscious team has accounts with least privilege available i.e. no root access. Please don't discount this critical vulnerability.

Why does your comment here contain more information than the blog post from tailscale itself?

I backup my different devices to my NAS. The user is a restricted backup user seeing only their own directory. I obviously don't want my torrent computer gets root on my NAS.

I personally use openssh, but someone else may use tailscale SSH.

Thank you for such a detailed explanation! I was wondering why we don't see MD everywhere and things like protein folding are all based on ML.

Is there anything like a getting-started for working with MACE or orbmol, a 'hello world'? Like maybe simulating Allolactose binding to lac1 ?

Heh. Unfortunately computer science is really the weird one out where open source is the norm. A lot of computational chemistry knowledge is mainly transferred in research labs and through college classes, and you have to pay for a lot of the tools. But hopefully I can at least point you in some useful directions. Here's a bit of an overview of the CC workflow and free options (note I haven't tried all the free options):

1. The first thing you need to do is start with some atomic structure. You can often find pre-existing pieces (through websites like pubchem), but you need to cobble them together. I use chemcraft and gaussview for this, but those are paid options. I've heard good things about the FOSS program Avogadro though.

2. Once you've built the structure of interest, you want to calculate properties of it. Note that we only infrequently simulate the atom's movement, since that's quite expensive. There's a lot of techniques that let us get the most bang for buck, like calculating transition states, or states of lowest energy which we call intermediates, or running something called an IRC, which follows the molecule to its lowest energy point (note we pretty much always think about these things in terms of energy). The traditional programs I use for this are Gaussian and ORCA (ORCA is free but you do have to create an account on their forum). Newer methods include MACE and orbmol, which are pretty much only used for trajectories (e.g. running through time) as far as I know. This newer crop of programs are much more FOSS, but they aren't as rigorous so my lab uses them only as a "rough draft". I pulled in one of my current notebooks if you're curious to try it out: https://github.com/smj-edison/qc-tools/blob/main/examples/ml...

3. Interpreting results/debugging problems. This is currently done in my workflow using a python script to extract values from the job runfiles (oh yeah, did I mention the terrible DX? A lot of this stuff is done by hand-writing a job text file and then using regex to parse the logfile). We do a lot of analysis based on relative energies (this is something called transition state theory of reactions).

I definitely glossed over a lot of details, but this is long enough as is. Definitely let me know if you have questions! If you do want a "hello world" though, I'd definitely check out the notebook I linked to. You'll have to create a file called "start.xyz" with some sort of chemistry program (or use AI to pull one for you), and then you can run the system through time. I would use a system I'm researching right now, but I'm not sure what our lab's policy is on broadcasting things to the world.

This is beautiful! I would love if we could model out a whole city, ideally using a much cheaper model.

I'm really curious how GML-5.3-flash would do. Very affordable, and it seems to do pretty well with 3D modeling.

yes experimenting with it actually, will update here! in fact we've generated most of the tourist spots in sf, should be reflected in the repo soon too

Hell yeah, whats your thoughts on doing it for like a whole city?

Yeah, I think this is fair criticism, and I definitely felt the pointedness in the tone as well.

On the overly-literal / narrow thing, I think thats the culture around evaluating predictions overall. Like, all those posts around christmas where people make predictions and evaluate how last year went. The rigor is the norm.

I'm not familiar with the culture, but that sounds reasonable.

I'll quibble with the term rigor. I think the article contains the strictness that word implies, but not the thoroughness.