It makes sense that there's some discovery mechanism - since Google loves to use it to prefer Chrome, GMail, etc when you're in one of their apps. I wish that there were more restrictions though where you only get implicit permission to query from apps that have the same developer ID. Maybe a mutual allowlist that has to be formed, or some sort of privileged intent where you at least have to tell Apple what's going on and that gives them some contractual right to sanction you if you're using it for nefarious purposes instead.

[1] excluding the clipboard copy count, that was novel!

Nevertheless though, we're blessed with a codebase that is several orders of magnitude larger than what the human brain can fully internalize so if you're trying to do anything remotely interesting, you're going to have questions.

I'm aware that the defective code was not written by AI but nonetheless, GitLab is what stands between many small organizations and their most precious resources. I was fortunate that 2FA stopped the damage, but what's going to happen the next time? What if my organization is permanently damaged because we taught the machines to go fast and break things, too [1]?

[1] VPN is an option but we're a non-profit with a number of non-technical users, so admittedly we're caught in a balance between making it harder to do things. As much as WireGuard is awesome, there's still a barrier.

They actually provided some example CVEs that wouldn't have been possible if in addition to containers, they were also using user namespaces https://github.com/kubernetes/enhancements/tree/217d790720c5.... The first example talks about "CVE-2019-5736: Host runc binary can be overwritten from container. Completely mitigated with userns." So it seems like getting root in a regular container gives you more of an attack surface, but if user namespaces are deployed, then it's even harder to do anything useful with it. I am looking forward to the aforementioned writeups since user namespace escapes usually mean another kernel bug.

That being said, it's pretty common for non-containerized processes to drop permissions to a low-privileged service account (like nginx running as `nobody`), so it definitely thwarts defense-in-depth in those setups.

In containerized environments, my understanding is their use of namespaces means you still need something more clever than just "patch out the authentication logic in su via the page cache" to escalate permissions in the system to break out of the container. That doesn't mean it's impossible in these exploits (the original copyfail writeup alluded to a second writeup coming to this effect - distinct from dirtyfrag though), but it does mean you're not going to be able to just spam the PoCs floating around.

As a side note though, I recently have tried to turn CSP on a website I run and the amount of garbage I see in the reports is astonishing. There's some noise from things like OpenDNS intercepting YouTube or Social embeds for people using the work-friendly or family-friendly options, but the sheer amount of things attempting to phone home to random URLs and random extension scripts injecting ads into the site would astonish you. My mental model of "toolbar hell" from the Windows XP days being gone has completely shattered.

The bulletpoint as-is just says:

> AI-resistant challenge: As we identify potentially fraudulent behavior from agents, we enable application providers to deter and mitigate malicious requests by requesting humans to be in the loop using the new QR code-based challenge. This AI-resistant mitigation challenge to prove human presence is designed to make automated fraud economically unviable.

Followed by

> Existing reCAPTCHA customers are automatically Fraud Defense customers, with no migration required, no action needed, and no change to pricing. Your existing site keys and integrations remain exactly as they are today.

It is probably me being a literal reader but "we enable application providers to deter and mitigate malicious requests by requesting humans to be in the loop" feels like it can be read as "Good news: by using reCAPTCHA, we're now interfering with agents that can solve the regular challenges" or "there's now a flag the application developer can set". This is the difference between me swapping off reCAPTCHA ASAP or just editing my configuration. I have to imagine someone somewhere anticipated the kind of reactions a number of us are collectively feeling (I too don't want to use my phone to browse the web more than I already do) and it feels irresponsible to publish a feature announcement without covering basic information like this for site administrators. Maybe they thought the second line about existing reCAPTCHA customers being moved over clears this up, but "Your existing ... integrations remain exactly as they are today" feels like again, literally, you won't have this new attestation requirement being presented to your users... but then why am I Fraud Defense customer!

The whole hour talk is worth a watch, even when passively doing other stuff. It is a neat history of Solaris and its toolchain mixed with the inter-organizational politics.

YouTube link: https://www.youtube.com/watch?v=-zRN7XLCRhc

Direct link to lawnmower quotes (~38.5 minute mark): https://youtu.be/-zRN7XLCRhc&t=2307

Edit: also to be somewhat objective, Instagram also offers a time limit. However, I've found that just by exiting the app (or it getting killed in the background), it basically clears the lockout so you don't even have to make the effort to click the "ignore" button.

What can the good guys do? Fire up Claude to improve their systems? Unless you have it working fully autonomously to counter-act abuse, I don't see how you can beat the "bad guys". There may be some industries where this is a solved problem (e.g. you can do all the validation server-sided, religiously follow best practices to prevent and mitigate abuse), but a lot of stuff like multiplayer video games will be doomed unless they move to a "you must use a locked down system we control" model. I honestly don't consider it liberating as someone that has various hobby projects, that now in addition to plain old DDoS I'll also have people spin up layer 7 attacks with just their credit card. It almost makes me want to give up instead of pushing forward in a world where the worst of the worst has access to the best of the best.

^1 https://docs.kernel.org/admin-guide/syscall-user-dispatch.ht...

Their big requirement is you need to not be doing any DNS filtering or blocking of access to what it wants, so I've got the pod DNS pointed to the unfiltered quad9 endpoint and rules in my router to allow the machine it's running on to bypass my PiHole enforcement+outside DNS blocks.

^1 https://wiki.archiveteam.org/

^2 https://wiki.archiveteam.org/index.php/ArchiveTeam_Warrior

^3 https://wiki.archiveteam.org/index.php/URLTeam

Comments URL: https://news.ycombinator.com/item?id=46968445

Points: 23

# Comments: 4

I'll take my turn on the soapbox to say I hope people keep posting about their adventures and misadventures in trying something new. I'd much rather be reading that than seeing yet another post on LLM-based agentic startups or pelicans riding bicycles.