Such a click-baity name. Open challenge to anyone to use this framework and reverse engineer the internal APIs iPhone uses to talk to Siri or shortcuts or app Intents.
There's no obscurity anymore. Best you can do is keep as much as possible server side. Curious if anyone has experimented with LLM's RE'ing obfuscated code. I suspect they handle it just fine.
> Curious if anyone has experimented with LLM's RE'ing obfuscated code.
Sort of, albeit I haven't checked with SotA obfuscation, I've seen GPT 6.1 Sol find a URL hidden under a xorpad + 2 layers of encryption, and I've seen Astra RE an AES whitebox.
More importantly, remember these game decomps (e.g. Pokémon)? Well, with minimal guidance and IDA MCP these new models are able to do this _and_ properly name all function/params/locals in a single day.
RE was arguably the first* task category to become completely solvable by LLMs on cybersecurity CTFs. This is exactly about obfuscated code, LLMs seem to have no problems with it now. The only probabilistic defense is to include tons of red herring so agent goes the wrong path and ends up in some tarpit like bruteforcing unrelated un-bruteforceable hashes.
* Cryptography category also went down fast but I think LLMs solved RE a bit faster. RE and crypto have in common that the task is completely local, but in classic RE tasks it is also usually ~obvious what you are supposed to do, you just need to do it - which agents do really well.
today i read that for new game studios, the possibility of going stream only (like amazon luna) instead of downloadable is a probable future if they want to prevent piracy.
It seems both very lossy and very expensive (even in a future where agents are cheaper to run) to have an AI clone, say, a 3D video game. It also seems borderline impossible to have an AI usefully clone a service for which the served data is the valuable part, like TikTok or Facebook.
Brother, and I say this with all the respect in the world. You are living under a rock (at least a digital rock) take a look at the RE of games or the stuff people are doing. Cost is not an issue for these thing. Groups of people pooling together their token budgets to do this.
I was replying to a hypothetical world in which 3D games were hosted on a Stadia-like platform exclusively. Thats hard to copy for an AI, because the work of doing so is very slow and expensive even assuming technology improvements.
It’s easier to roughly copy, sure. As you say, examples of that exist. But outright cloning a game with only streaming/realtime content available? No decompilations, 3D asset parsing, or ability to run lots of simultaneous game loops? That would take ages and/or break the bank.
I’m not saying AIs are technically incapable of doing this. Rather, I’m saying that Stadia-as-DRM makes the act of copying technically infeasible (costly, slow).
I don't know. I want to say "no", and I think you want me to, too. But I've lived awhile and seen a lot of extremely accessible free (as in beer, as in freedom, as in "nobody will know" stolen, as in free-with-blockable-ads, and so on...) software rise and fall, and I've seen many payware alternatives (sometimes outright inferior ones) remain competitive long after your logic says they should. So ... I don't know.
I ported a linux game binary from a Cortex-A7 to a Allwinner A133P (32bit to 64bit) in about two hours using Claude over the weekend. (I didn't have the source). SDL and the games encryption were the only things that made it more than a 15 minute activity.
If it has a binary value you can aim the Ai at, (aka does it work or not), the models are able to figure it out.
For laughs I'm probably going to have a Qwen model on my Mac min with the Pi harness do the same task in a week or two.
Hmm, prolly no reason this wouldn't work on older educational dos/windows games and porting them to linux. Damn, another weekend is going to be lost isn't it.
25 comments
[ 2.4 ms ] story [ 39.6 ms ] threadSort of, albeit I haven't checked with SotA obfuscation, I've seen GPT 6.1 Sol find a URL hidden under a xorpad + 2 layers of encryption, and I've seen Astra RE an AES whitebox.
More importantly, remember these game decomps (e.g. Pokémon)? Well, with minimal guidance and IDA MCP these new models are able to do this _and_ properly name all function/params/locals in a single day.
* Cryptography category also went down fast but I think LLMs solved RE a bit faster. RE and crypto have in common that the task is completely local, but in classic RE tasks it is also usually ~obvious what you are supposed to do, you just need to do it - which agents do really well.
Obfuscation is no obstacle at all in my experience.
It’s easier to roughly copy, sure. As you say, examples of that exist. But outright cloning a game with only streaming/realtime content available? No decompilations, 3D asset parsing, or ability to run lots of simultaneous game loops? That would take ages and/or break the bank.
I’m not saying AIs are technically incapable of doing this. Rather, I’m saying that Stadia-as-DRM makes the act of copying technically infeasible (costly, slow).
If it has a binary value you can aim the Ai at, (aka does it work or not), the models are able to figure it out.
For laughs I'm probably going to have a Qwen model on my Mac min with the Pi harness do the same task in a week or two.