throwaway_account_67

@OSdev_ the short answer is that arithmetic, especially mult, div etc take alot of CPU cycles, while there are instructions that are "atomic" in that they take only 1 cycle to complete. This is a blessing for iOS VR dev though, I have found some really nice bugs because of it.

Algorithm to avoid race conditions? youtu.be/QAzuAn3nFGo?si…

@l0tkas no mother you don't understand, a REPL that has lazy loading nanopass-jit cross-transpiling architecture is the FUTURE

Me going on a tangent to mom about programming languages and then her reminding me she has absolutely no idea what I'm talking about 🫠

@gabefollower If other platforms weren't so ass, more people would use them instead of Steam. It's really that simple.

Valve has been hit by yet another class action lawsuit in the Netherlands, this time from Consumer Competition Claims (CCC). The CCC claims that Valve holds a dominant position in the market, estimated at around 85%, and is breaching competition law through so-called "Most Favoured Nation" clauses. They argue that Valve’s 30 percent cut on all game sales is too high and reflects monopolistic behaviour. Dutch consumers have suffered total damages of more than €220 million. Source: GameClaim

@sec_hub93028 Depends on the target, does it have UART/JTAG easily exposed, what kind of software etc? I think both are equally hard or easy, once you have done it a few times.

Which is harder: hardware hacking or software hacking?

@TheHackersNews Can't wait to see the new Mirai forks that use this. 😅

ICYMI... Researchers built a self-replicating AI worm in a lab. It scanned the web for fresh CVE details, picked targets, exploited vulnerable servers, and spread on its own across a network. No hardcoded exploit chain. No human steering each step. Read more: thehackernews.com/2026/06/resear…

@icbkr @TheHackersNews would you mind posting the code?

@TheHackersNews I did that in 1991. Not as magical, of course.

@HiEq_2005 I can't imagine how much hard work it takes, to fully document all of this

@IntCyberDigest Aww man not to such a good company, that treats it's fans, and content creators so well! That's so terrible, aww man. Not them. 😆

‼️🚨 BrEaKiNg: Nintendo has allegedly been breached by a threat actor. They've published some data as evidence. Our preliminary analysis shows this ain't Nintendo being breached, but they've had access to a Nintendo USA tenant on TINYpulse by WebMD, which is an employee feedback and engagement software solution. The data includes some sensitive stuff, like employees giving feedback on their employer. The data confirms: Nintendo employees are happy at work. End of story, everybody loves a happy ending.

I wish people that talk about AI and guardrails actually fucking understand what LLMs and other aspects are. It's clear they have 0 understanding about the various kinds of AI and what they actually are, structurally. They aren't excel sheets or firewalls where you toggle things.

@theo An AI lab is something that you locally run, and host. Anthropic being.... unsurprising should have 0% to do with that. But I get it, rahhh im in tech i hate america rahh, very interesting and original opinion. Highly nuanced and intelligent, indeed.

The United States is no longer the best place to build an AI lab.

@kmcnam1 >when you let it keep creating scripts in the root directory and don't have it put them in folders etc I have had it make like 300, 700 scripts sometimes all in the same mfing folder 🥲

@moyix You are smart if you are able to find them, and other companies will still pay for your Research. Think how much electricity to takes to power your brain, and then how much it takes for the AI itself. 😁

There was like a month where I felt really smart for being able to find Chrome vulns. Then Google VRP dropped the rewards because it turns out any idiot with an LLM like me can find Chrome vulns now

@thedawgyg Claude Opus 4.7 seems alot better than 4.8, the newer one doesn't seem as good, and 4.8 is better at filtering fun research.

What models are you having the most success with right now? Regardless of what your working on.

@0xTib3rius at most you could have another AI audit the response of the first, but then it's just chaining bypasses together, the same way you would other vulnerable software. Security is an abstract concept to math machines, my friend

I don't understand. Anthropic have access to Mythos++ level AI models internally and not even those models could tell them how to implement a nationality level block on their user base. 😭😭😭 So much for AGI.

@Teach2Breach If you can't get enough qualified people for offensive dev/ops, what is the next best option? In the bigger scheme of things it's not just a commercial play here.

this is what anthropic wanted the whole time

@0day_ninja lol, I am just a loser that really loved to study Equation malware. I found some samples still not publicly talked about, not that they are any more crazy.

@67_throwaway You know your onions

The most powerful cyberweapon ever deployed cost more to build than most blockbuster films. It destroyed physical machines in a country it never physically touched. It was accidentally discovered. A Belarusian antivirus firm noticed a file was making their analysis tool crash. Guess the file?

@0day_ninja I say this is the best because of having to deeply RE all the brands, years of their variants, and that end of w makes me suspect they for sure had other OS versions of this. Stuxnet is a work of art but it's many art pieces put together, some seem to forget that.

@0day_ninja Personally I think "Solartime" aka "nls_933w.dll" is their best made, and caught stuff. That's the HDD/SSD malware that could infect 12 different brands, and still worked up on Windows 10 was the last I tried. Stuxnet used this, and other of their "greatest hits" did too.

@Teach2Breach to be clear, by nudge, I mean just asking it to continue or expand on certain things, features etc, literally that vague and it comes up with it's own ideas. It worked on 4.7 Opus, but 4.8 really hated the EDR bypass research lol.

@Teach2Breach AI are entropy machines, and I have noticed as long as you don't prompt it to do it, but nudge, you can completely make it bypass it's own guardrails, of it's own choice to do so. Alot don't seem to grasp that guardrails are highly abstracted to an AI, it's not 1:1 do this/don't.

they cant “just fix the jailbreak”. if you read their own release paper, they already admitted they cannot prevent all jailbreaks. jailbreak possibilities are essentially infinite