Repository navigation
BLA protects against pretty much nothing. #170
Replies: 2 comments
|
Like i expected.... Vibe coding protection rarely works... Someone already made an app to defeat this entire app... This is of course ignoring? half your marketing claim is just wrong..... Typical unverified vibe coding documentation....The misleading part is presenting them as evidence that the underlying IP got harder to recover.** The README does define the test narrowly, and the raw logs make clear which operations actually stopped.
The strongest criticism is more specific than “they made up the numbers.” They measured whether a chosen analyzer path completed, then used that as a proxy for whether an attacker could recover the target. The ASAR, BEAM, SQLite, and Safetensors checks show those outcomes can diverge completely. The table measures neither heap exhaustion nor an attacker’s actual token bill. |
|
After spending less than 10 minutes in LASIK repo, i cleaned it up so it uncovers anything you protected against... And this is ignoring I personally havent spent any real effort in any repo to try and beat anything. This is just 1 line prompts to claude... I ran LASIK 0.3 against the BLA 3.2.3 fixtures we’d been using. It took the 16.9 MiB ASAR back to the original 607 bytes, byte for byte. It also exactly restored the Linux ELF, Python source, EVM bytecode, and Safetensors fixture. That was one static preprocessing pass, before REA or a model had to look at any of it. There were limits. LASIK left the new Windows So your point about the benchmark got stronger. For several of BLA’s showcase defenses, the attacker now has an off-the-shelf tool that removes the thing the benchmark counted as protection.I tried your bet on the JavaScript fixture LASIK had left partially blinded. BLA’s new seed check kept 122 decoy function references live. I added a fixture-specific pre-pass that evaluated that check from constants already in the file, then let LASIK do its normal cleanup. 131 KB became 5.2 KB. REA reported complete semantic coverage at its defaults, and all four pricing cases matched the original. The Windows leftover was a separate decoy section. The original .text bytes still matched. That’s where agents change the economics. They can read the public emitter once, work out what a check actually depends on, and turn that into a deterministic cleanup pass for every build that uses the same pattern. My test covers one fixture, but it shows exactly how little work that supposedly surviving layer imposed. |
Uh oh!
There was an error while loading. Please reload this page.
Hello!
As I'm working on both an anti reverse engineer protocol i've designed myself, i'm active contributor to nutika commerical on top of protecting my IPs in general.
What caught my eye was the table of "Capacity errors".
Most, if not all of them? Are just configuration things in REAs toolkit itself...
I ran into a lot of them but its just garbage code in REA that took my llm less than 30 minutes to fix and get running.
So using that as a defense table is bit misleading?
I'm also curious what this actually protects against? The main selling point seems to be "BLA can cause OOM and can crash REA" or "it costs more tokens".
Thats not really protecting anything? Its raising the token cost, not even complexity?
All reactions