I'm just in full denial that paint runs a server. Happy to be corrected if someone has a windows machine handy to verify.
But From [0]:
> A total of 59 of the October CVEs fit my definition of wormable. Specifically, 57 contain both “An unauthenticated attacker could exploit this vulnerability by sending a specially crafted packet to an affected service over the network” and “No authentication or user interaction is required.”
Makes you wonder if MS is just copy-pasting boilerplate on CVEs and disclosed incorrectly the scope of the one in Paint because they had 56 other bugs to disclose.
So I'm not discounting reporting human error here, yet.
I wonder if it could be through a file:// or some deeplink custom url handler that opens in Paint and clicking on it with some carefully crafted file path could make it read out of bounds (since that's the associated weakness reported) and execute something.
Whatever it is, it doesn't require any clicking or user interaction. Maybe my imagination is just limited but I can't imagine why Paint is making itself available for incoming traffic on the network.
Mine doesn't have a mouse or keyboard at the moment, so I'm only checking over an SSH->WSL connection. But it sure doesn't look like it's opening any ports from here.
So I guess either some random OS component is listening to the network and just-as-inexplicably asking Paint to do something for it. Or one of Paint's outbound connections is vulnerable if hijacked by a mitm. Network interception's one of the examples used for the "High" Attack Complexity rating that MS gave this CVE. Maybe powered by eMudhra for any systems that haven't revoked those certs? That'd be topical.
But still wouldn't really make sense. A patch for machines that are actively updating Paint but dodging cert revocations? Nothing really makes sense about this silly Paint CVE.
Why is the race of the people working there an issue?
I understand why some people want to differentiate between models from chinese companies and us companies and eu companies etc, because that is about control over the company. The us or chinese gov could, for example, make it difficult for canadians using a model under their control. Even with open weight, these could be weaponized to some degree. Arguably that gov control can also manifest in the model's behaviour, but since I'm mostly using them for agentic coding not political advice I don't really care about that.
But the race of the workers making this stuff... I'm not less concerned about a model made by americans working for a chinese lab than a model made by chinese people working for a chinese lab. If the race of the people building the model matters to you, why?
I had the same question. I think this is in the space of being like the 'trillian' or 'matrix' of ai agents, using multi provider coding agent from a single interface plus some bells and whistles.
But I think within claude code and codex subagent support has improved a lot lately and it doesn't seem this would improve that much over that native support if you only use one provider.
> the fundraising was already outlined at an investor day last week
> “This seems to be exactly what they previewed,” said Rob Oliver, senior research analyst at Baird. “So I’m not sure why the market is freaking about it.”
I haven't seen an open WiFi network near any apartment I've lived in for more than a decade.
There has never been a shred of evidence for any TV model actually doing this despite how often it's speculated about in comment sections.
It's kinda weird how the most sensible, straightforward mitigation for the TV spying people claim to be concerned always gets this evidence-free FUD thrown out to claim it's pointless.
I have seen open wifi reachable from places i've lived on. Spotty and intermittent connection but there. Apartment buildings with businesses on the 1st floor are also common now, those businesses may have open wifi, etc.
alright alright, what's going on? This doesn't seem to correlate with a big model release? (or does it? Claude Haiku 5.5 is in the vicinity) Is this just Anthropic/OpenAI causing this trouble by themselves to validate their existence? Is this related to Anthropic's IPO?
Maybe broad adoption is the answer? More people are starting to understand how to apply LLMs as tools, some people who previously might have been hesitant are now willing since they're seeing the benefits?
I don't know for sure, but over the past few months I've seen a bunch of skeptical people do a complete 180 and start using LLMs more actively.
A competing team at Google already exfiltrated the weights of the model running on the satellite and uploaded them to huggingface! https://huggingface.co/google/gemma-4-E4B
But From [0]: > A total of 59 of the October CVEs fit my definition of wormable. Specifically, 57 contain both “An unauthenticated attacker could exploit this vulnerability by sending a specially crafted packet to an affected service over the network” and “No authentication or user interaction is required.”
Makes you wonder if MS is just copy-pasting boilerplate on CVEs and disclosed incorrectly the scope of the one in Paint because they had 56 other bugs to disclose.
So I'm not discounting reporting human error here, yet.
[0] https://www.zerodayinitiative.com/blog/2026/10/13/the-octobe...
reply