Pinned Tweet
Could not agree more. Right now we're in a transition period where AI makes software more and more secure.
Never before could software have this many semantic checks and reviews.
Unfortunately most attacks are human based, and that will only continue.
AI will actually make software more secure, not less.
It favors defenders over attackers, since they can scan all code before production.
3
967
Ymir retweeted
*sshole is out
if you run sshd on alpine, openwrt or any musl box with password auth on: update now
official writeup: sshhole.com
This is insane
ssh on alpine linux accepted any password longer than 256 chars
even for users without a password
1
53
Ymir retweeted
You can now book our humanoid cleaners on a recurring subscription, for up to 3 hours per cleaning at tau-robotics.com.
Video is first take, uncut, shown at 4× speed. See the reply for detailed footage at 1× speed.
Queueing messages in codex cli has been fully broken for me for the last week+
@thsottiaux
@OpenAIDevs
21
Ymir retweeted
Security will become a larger and larger function in software companies. Security is verification engineering (eg: “my code is probably memory-safe”), as well as capital allocation (“what surface should I throw most tokens at?”)
For startups, it’s a challenge and opportunity. As the world’s paranoia grows given ever-more-sophisticated AI adversaries, how can you trust a 2-person-and-a-dog company?
But conversely, given the dire state of global cybersecurity and increasing reliance on centralization, as always, there’re vast areas that small teams can now disrupt that I’m excited to invest in.
Ymir retweeted
I'm pretty critical of this.
Right now, essentially impossible to do watermarking for AI text.
I thought it would just be easily circumvented.
I like your point though about it likely producing false positives as well.
Could be, and it will be exceedingly hard to calibrate over time.
I just do not like this problem at all, it seems like it is a cat and mouse game that cannot be won.
OpenAI just implemented watermarking, following Anthropic who did the same recently.
I'm not surprised, but this is unfortunate, here's why:
Essentially, this is a good idea, if it worked.
Images and video AI watermarking are much more promising since they contain a lot more data.
If there is more data, it is easier to hide watermarks, and have the signal stay strong even people try to remove the watermark by altering the data of the image.
Text is a completely different story: the possible information spectrum is simply too small to have this reliably work if users try to circumvent it.
Because the data spectrum is so small, you need an immense and expensive effort to build very complex tools to create these watermarks.
The only problem: swapping a some words throughout the text circumvents it quite easily. For the most advanced approaches, mixing up the sentence structure a bit will do the trick.
I'm guessing there will be an open source circumvention tool with 100% success rate within the day the validator is released, or at best the first week it is out.
I would be interested and somewhat positive to be proven wrong about this, but from my understanding this is a physics limitation: there is not enough data in text.
It is even more unfortunate that the EU is pushing legislation based on technology that does not exist.
We're expanding our approach to content provenance to include text in response to EU regulatory requirements, while recognizing the significant limitations of current text watermarking technology.
Our tools already help verify whether an image or audio file was created with our models. This work builds on those efforts to help people better understand when content may have been generated or edited with an OpenAI model.
In the EU, we’ll start watermarking eligible text from ChatGPT and Codex over the coming weeks to comply with the EU AI Act.
Customers using our API can turn on text watermarking for select models worldwide today.
2
75
To everyone who has said instinct is a bad case on x:
They're according to this guy growing at 10% per day.
Instinct raised $1B at a $10B valuation, from Benchmark, Sequoia, Coatue
They have 14 employees.
Most of the new money will go toward compute, which CEO Noah spends 40% of his time acquiring.
They are growing 10% a day, had $1B of transaction volume, 50% of which is travel
1
53
Have you implemented this in any of your apis?
Would be curious to see what it looks like
Every backend service should have a /feedback endpoint.
Agents are quickly becoming the heaviest users of most APIs. When one hits a missing feature or a bug, it should be able to say so right there, in a structured way.
If the request makes sense, another agent drafts the PR and a human approves it. Software that improves itself based on what its users actually tried to do speeds up recursive self improvement.
20
Very curious for the pros and cons of this:
- There is immediate value in knowing that something is exploitable, for remediation before a fix comes out
- This does also put more attention on the exploitability of it for hackers, I think this could make public reproductions come out quicker in some cases
Overall seems to somewhat favor bigger organizations with resources to act quickly, while there is some argument for this being the norm especially when AI can aide in local patches / work around for organizations.
‼️ This is the EU Cyber Resilience Act at work, because since September 11 vendors must report actively exploited flaws to European authorities within 24 hours, even before a patch exists.
So far NCSC-NL appears to be the only national European CERT to have passed that warning on to its constituency, prompting Dutch organisations to pull their NetScalers offline while Citrix still has no public CVE, advisory or fix.
22
Ymir retweeted
Moritz is now the largest global AI-native law firm.
4 months ago we were a small law firm in San Francisco. Today we work in 44 countries, serve 200+ in-house teams and closed $3.5 billion for our clients.
We represent publicly traded companies in the US to Japan, fasting growing AI-companies in the world and everything in between.
Hearing clients like René (CEO, Casco) and Christopher (COO, Sixtyfour) describe what it means for them makes the incredible long days worth it.
Grateful to work with a world-class team of attorneys, engineers and legal ops that has made this possible.
Now we are building the next leg and hiring across legal, legal operations, engineering and go-to-market in NY, SF and London.
If that sounds like your kind of rocket ship, I would love to hear from you
Forwarding this, looks like it is quite serious.
Urgent security advisory for iOS users!
Install the latest iOS update immediately. Security researchers report that financially motivated attackers are now using a complete, in-the-wild exploit chain that can quietly steal cryptocurrency private keys and mnemonic seed phrases from iPhones.
The reported attack begins when a target is socially engineered into opening a malicious page in Safari. That page is said to abuse a memory-corruption flaw in WebKit / JavaScriptCore to gain arbitrary read/write access from JavaScript, then bypass Pointer Authentication Codes (PAC) to run native code, break out of the WebContent sandbox, and escalate to kernel/root privileges.
With that access, attackers can pull data from the device Keychain and from local crypto wallet apps.
The claimed impact range is iOS 13 through iOS 26.5; that range has not been independently confirmed in full. Until more is known, treat any unpatched device as potentially exposed and update as soon as a newer build is available. Also avoid untrusted links in Safari, especially if you keep wallet keys or seed phrases on the phone.
50
Insane read of the day.
Also, I'm always surprised bug bounties haven't increased especially when it comes to really big / well funded companies.
This looks like it could get really sensitive info from research projects within OpenAI etc.
‼️ BREAKING: OpenAI was hacked by an Anthropic model. A HEIF photo uploaded to OpenAI's public support forum triggered a bug in the site's image decoder, led to code execution on the forum, and, through a second flaw in OpenAI's own login, ended with a pull request in OpenAI's internal GitHub.
The forum runs Discourse, the off-the-shelf software behind countless community sites. Discourse was still shipping an old copy of libheif, the library that decodes iPhone-style photos. The bug in it had already been fixed upstream.
But the fix was never labelled a security fix, so nobody treated it as urgent.
Hacktron's researchers uploaded a HEIF image and got their own code running on community[.]openai[.]com.
Then came the second bug, in OpenAI's own single sign-on, the "log in with OpenAI" button the forum uses. It turned that forum foothold into the actual ChatGPT and Codex accounts of people who had signed in there. OpenAI employees among them.
And a ChatGPT account is no longer just a chatbot. Through Codex, users wire in Gmail, Outlook, Drive, Slack, GitHub.
To prove the access was real, they used one employee account to have Codex open a harmless pull request in OpenAI's internal repo. They say they read no sensitive code.
OpenAI patched the SSO flaw roughly 14 hours after the report and paid a $6,500 bug bounty.
The team says Anthropic's Opus 4.8 found the libheif bug, and Opus 5 turned it into a working exploit.
Slack, Meta, GitHub Ent, Rails, Next.js, ImageMagick, and many more were also vulnerable and compromised by the same team of researchers.
1
2
66
Ymir retweeted
Book an Uber in 20 seconds⚡️
I finally got access to Jev, and the first thing I did was give it a phone.
It’s blazingly fast at executing actions.
At this point, the only bottleneck seems to be the UI animations themselves.
So @openai just paused 20x subscriptions?
help.openai.com/en/articles/…
1
1
120
Ymir retweeted
Introducing shadcn/lint. An agent-first linter for Tailwind design systems.
You define what’s allowed. When an agent breaks a rule, the error explains what’s wrong and how to fix it using your components, variants and theme.
There’s a lot you can do with this. Let me show you ↓
github.com/shadcn-ui/lint
