Can ChatGPT, Claude, or Copilot reverse-engineer code protected by JavaScript Obfuscator?

Maximum-mode protection emits a per-build polymorphic decoder. Every release regenerates the decoder shape, key derivation, identifier prefix, and constant-pool encoding. LLMs work by pattern-matching against known transform signatures, so an AI deobfuscator that solves one build cannot apply the same approach to the next.

Is the original source recoverable from the obfuscated output?

No. Once the symbols and comments are transformed, the original source cannot be recovered from the obfuscated output - not by hand, not by automated deobfuscators, not by LLM-based tools.

Can JavaScript Obfuscator protect JavaScript embedded in HTML, PHP, ASP, ASPX, or JSP files?

Yes. The desktop app protects JavaScript embedded in mixed files including HTML, PHP, ASP, ASPX, and JSP.

Is anti-LLM Maximum mode included in the lower-priced plans?

Yes. Maximum-mode anti-LLM protection is available in every paid tier starting at $29/month, not gated behind an enterprise upgrade.

AI & LLM-Resistant Protection · Since 2004

Anti-LLM JavaScript obfuscation, in every paid tier.

Maximum-mode output regenerates its decoder shape, identifier prefix, and constant-pool encoding on every build. ChatGPT, Claude, and Copilot have no fixed signature to pattern-match. Published pricing from $29/month — no sales call required.

Try It Free Online $29 starts at 1 GB/mo Can ChatGPT reverse this? ›

Anti-LLM in every paid tier Per-build polymorphic decoder. ChatGPT, Claude, Copilot — no fixed signature to learn.

22 years of releases Continuous shipping since 2004. Built by Richscripts Inc, Toronto, Canada.

10× bytes-per-dollar 1 GB at $29 vs. ~100 MB at the comparable $19 tier elsewhere.

Live Preview

Pick a protection level. See exactly what the output looks like.

Same input, three different transforms. Maximum mode is what AI assistants can't pattern-match.

Standard Balanced MaximumAnti-LLM

Input

function validateLicense(plan) {
  if (plan === 'trial') return false;
  return plan.expiresAt > Date.now();
}

Standard · rename only

function _0xa1(_p){
  if(_p==='trial')return!1;
  return _p.expiresAt>Date.now();
}

Input

function validateLicense(plan) {
  if (plan === 'trial') return false;
  return plan.expiresAt > Date.now();
}

Balanced · strings + dead code

var _s=['trial','expiresAt'];
function _0xa1(_p){
  if(_p===_s[0])return!1;
  return _p[_s[1]]>
   Date.now();
}

Input

function validateLicense(plan) {
  if (plan === 'trial') return false;
  return plan.expiresAt > Date.now();
}

Maximum · LLM-resistant

(function(){var _0xa3=_dec(0x4a);
function _0xa4(_p){var _st=0;
 while(_st!==-1){switch(_st){
  case 0:if(_p===_dec(0x4b))return!1;
   _st=1;break;
  case 1:return _p[_dec(0x4c)]>
   Date[_dec(0x4d)]();
}}}})();

Per-build polymorphic. Decoder shape, key derivation, identifier prefix all regenerate every release.
Encrypted constants. Strings and numbers exist only after runtime decode — LLMs lose their anchor points.
Flat control flow. if/else replaced by state-machine dispatch — structural cues gone.

2004 First shipped. Continuous releases for 22 years. Heritage

$29 = 1 GB Entry tier with anti-LLM Maximum mode included. Pricing

Per-build Polymorphic decoder regenerated every release. Anti-LLM

4 paths Online, desktop GUI, hosted API, npm CLI. Workflow

The 2026 Question Buyers Are Asking

Your old obfuscator may be one ChatGPT prompt away from useless.

AI assistants pattern-match against transform shapes they have already seen. The fixed-output obfuscators that were good enough in 2020 are now training data. Here's how Maximum mode is structurally different.

Static obfuscators are training data.

Open-source javascript-obfuscator output is in every modern LLM's training corpus. Models have learned the inverse transform and can apply it to any release that uses the same fixed shape.

Per-build polymorphism breaks the pattern.

Maximum mode emits a different decoder shape on every release. The LLM that "solved" version 2.0 sees a structurally different program in 2.1 and has to start over. It can't accumulate progress.

No anchor points to reason from.

Encrypted string pool, flat control flow, identifier prefix unique per build. The literals an LLM normally uses to recognize libraries, API calls, and intent are all gone — replaced by opaque indices into a runtime-decoded pool.

Read the full technical analysis ›

Find Your Fit

Pick your threat model. We'll point you at the right preset.

The right amount of protection depends on what an attacker actually wants from your code. Pick one and the recommendation appears below.

Tier 1

Anti-copying

A competitor scrapes my script and runs it on their own site. I want them to get a working file they can't easily modify or extend.

Standard preset →

Tier 2

Anti-IP-theft

Someone wants to understand my algorithm well enough to reimplement it. They have time, tools, and motivation. I want to make that effort uneconomical.

Maximum preset →

Tier 3

Anti-active-attacker

Someone wants to tamper with my running code — bypass a paywall, grant a feature, manipulate game state. Active attacker, real-time threat.

Maximum + monitoring →

Recommended: Standard preset. Variable renaming and string encoding stop the casual copy-and-paste copycat. The Free tier is enough — no credit card required. Upgrade only if your usage grows beyond 200 MB/month.

Open Standard Preset ›

Recommended: Maximum preset. Per-build polymorphic decoder, encrypted constant pool, flat-transformed control flow, self-defending wrapper. The threshold an attacker has to cross becomes uneconomical for most non-state actors. Available in every paid tier from $29/month.

Open Maximum Preset ›

Recommended: Maximum preset + a runtime monitoring suite. Maximum mode raises the cost of analysis; runtime monitoring catches tampering when it happens. Server-side authority on anything that grants access. Obfuscation is one layer in a layered defense, never the only one.

Open Maximum Preset › Runtime Defense Guide

Pricing

Published plans. Anti-LLM in every paid tier.

No sales call. No "contact us for pricing." Maximum-mode anti-LLM protection is included from the entry tier up — never gated to a higher plan.

Bytes per dollar at the entry tier

JavaScript Obfuscator

$29 → 1 GB / month

Comparable

$19 → 100 MB

10× the bytes per dollar. Anti-LLM Maximum mode included from $29.

Free

$0/month

A no-cost starting point for basic obfuscation workflows.

200 MB monthly quota
20 files per request
Name mangling
String encoding

Get Started Free

Basic

$29/month

Anti-LLM Maximum mode for individual developers shipping active products.

1 GB monthly quota
50 files per request
10 MB max file size
Compression included
All Free features

Get Started ›

Corporate

$49/month

The balance of capacity and advanced controls most teams pick.

3 GB monthly quota
1000 files per request
30 MB request size
Replace globals + protect members
Dead code insertion

Choose Corporate ›

Enterprise

$99/month

Higher limits for teams with larger release pipelines.

9 GB monthly quota
3000 files per request
120 MB max file size
Best compression ratio
All Corporate features

Get Started ›

Compare all four plans in detail ›

The Question Every Procurement Reviewer Asks

"Can ChatGPT, Claude, or Copilot reverse-engineer this output?"

Short answer: no, for output that regenerates its decoder shape every build. Yes, for static obfuscators in their training data. The mechanism that makes the difference: per-build polymorphism removes the fixed signature LLMs rely on.

We wrote the long-form technical answer — what AI assistants can actually do today, where they break down structurally, and the specific transforms (encrypted constant pool, flat control flow, polymorphic decoder) that defeat their pattern-matching approach.

Read the full answer ›

// What an LLM sees on a Maximum-mode build:
//   (function(){var _0xa3=_dec(0x4a); ... })();
//
// What it CAN'T tell you:
//   - what _dec(0x4a) decodes to
//   - which case the state machine runs first
//   - what the original variable names were
//   - what the strings are
//
// All of that lives behind the decoder,
// which only runs at runtime.

Procurement-Ready

Get answers without a sales call.

Source-handling, release validation, compliance vocabulary, and workflow-to-policy fit are all documented. Your security team can review without scheduling a discovery meeting first — and that's the point.

Open Security & Trust Source Handling Doc Release Workflow

✓

In-memory processing

Submitted JavaScript is processed in server memory only. Temporary upload artefacts removed after each request. Desktop workflow keeps source on the workstation throughout.
✓

Compliance vocabulary mapped

GDPR data minimisation, OWASP A04/A09, PCI DSS 4.0 client-side script integrity, HIPAA PHI-adjacent code, NIST SSDF release-protection — all addressed on one page.
✓

Honest about what we don't have

No SOC 2 or ISO 27001 today. We publish what we do have: source-handling behaviour, release validation, support channels, and the local workflow path for projects that need source to stay local.

Desktop App

Your source code never leaves your workstation.

For PHI-adjacent web apps, on-prem build pipelines, or any project where the JavaScript source code can't be uploaded to a third-party server — use the desktop GUI. Batch process entire projects, including JavaScript embedded in HTML, PHP, ASP, ASPX, and JSP. Generate a deterministic command line you can check into your release pipeline.

Download Desktop › Take a Tour ›

Three Ways To Start

Pick the entry point that fits how your team actually ships.

Try the playground in 30 seconds for an experiment. Download the desktop app for batch protection on PHI-adjacent or on-prem code. Or read the long-form technical analysis if you're still deciding whether anti-LLM protection is what you actually need.

Try Online Obfuscator Download Desktop Read the AI & LLM Article

Anti-LLM JavaScript obfuscation, in every paid tier.

Pick a protection level. See exactly what the output looks like.

Your old obfuscator may be one ChatGPT prompt away from useless.

Static obfuscators are training data.

Per-build polymorphism breaks the pattern.

No anchor points to reason from.

Pick your threat model. We'll point you at the right preset.

Anti-copying

Anti-IP-theft

Anti-active-attacker

Published plans. Anti-LLM in every paid tier.

Free

Basic

Corporate

Enterprise

"Can ChatGPT, Claude, or Copilot reverse-engineer this output?"

Get answers without a sales call.

In-memory processing

Compliance vocabulary mapped

Honest about what we don't have

Your source code never leaves your workstation.

Pick the entry point that fits how your team actually ships.