AI reverse engineering means using AI agents to read a compiled app and rebuild working source code from it, a job that used to take skilled specialists weeks or months. In November 2026, researchers at Intezer used AI agents to rebuild Slack’s desktop client, Apple’s Finder and a Russian state-linked malware sample. That makes copying software, auditing it and finding its flaws much cheaper for defenders and attackers alike.

Reverse engineering has always been possible. What’s changed is the cost. A security team that once needed a senior analyst and a month can now point an AI agent at a binary and get readable, runnable code back. The same week Intezer published its results, a developer released a tool built specifically to slow AI agents down, and a two-person startup released open-source “clean-room” rebuilds of Adobe’s best-known apps.

Here’s what actually happened, why it matters even if you never touch code, and what software makers and regular users should do about it.

Key takeaways

  • Intezer’s researchers used AI agents to rebuild Slack’s client, Apple’s Finder and a TinyTurla-NG malware sample from their compiled code.
  • The rebuilt malware matched the original closely: 1,034 of the original’s 1,041 functions matched, with 92.8% similarity.
  • Developer Timo Zhang released Blind LLM Agents, a tool that makes code harder for AI to read. He says it raises the cost but isn’t unbreakable.
  • Cloning commercial software is now cheap, which raises legal fights over copyright and “clean-room” rebuilds.
  • For ordinary users, the near-term effect is faster discovery of security holes, so updating apps quickly matters more than ever.

What is AI reverse engineering?

When developers ship an app, they usually ship compiled or minified code: instructions a computer can run but a person struggles to read. Reverse engineering is the process of working backward from that code to understand how the app works, and sometimes to recreate it.

Traditionally that meant tools like Ghidra or IDA, plus a lot of patience. An analyst would rename thousands of functions by hand, guess what each one did and slowly piece the logic together. AI reverse engineering hands much of that grind to language models, which are good at reading code, guessing intent and writing clean replacements.

The models don’t need the original source. They work from what’s on your machine, which is why the technique applies to commercial apps and malware alike.

What did Intezer’s researchers rebuild?

Intezer, a security company, ran three tests, described in detail by Help Net Security. A Codex agent drove the work and made the engineering decisions, with other models, including Claude and Kimi K3, handling specific jobs such as identifying libraries in Slack’s code.

Target What it is Result
Slack desktop client More than 500,000 functions, most from third-party libraries A working client plus a compatible server. The team then added an AI autocomplete feature Slack doesn’t have.
Apple Finder A 25.8 MB macOS executable Mostly rebuilt, with gaps around online-only files, sync status and features that depend on other macOS services
TinyTurla-NG A backdoor Cisco Talos links to the Russian Turla group About 3,100 lines of source. 1,034 functions matched the original, with 92.8% similarity at 98.7% confidence.

The researchers rated the malware the hardest of the three, and it’s the most striking result. A rebuilt malware sample that behaves almost exactly like the original is a gift to defenders, who can study it safely, write detection rules and understand its tricks. Intezer put the problem plainly: “the same method that opens up a commercial app also opens up malware.” Their conclusion: “The software industry will have to adjust.”

Why does AI reverse engineering matter for software companies?

Cloning gets cheap

If an agent can turn a compiled app into clean, maintainable source code, a competitor no longer needs years to copy a product’s features. The startup ArtCraft has released open-source tools that recreate the interfaces of Photoshop, Illustrator, Premiere and other Adobe apps, built with AI-assisted reverse engineering, according to The Week. Its founders call them clean-room reimplementations. Critics have warned of lawsuits.

The legal lines are blurry

A clean-room rebuild traditionally meant one team documents how a program behaves and a separate team, which never saw the original code, writes a new version. Whether an AI that has read the original code can count as a “clean room” is an open question. Software licenses often ban reverse engineering, but enforcing that against open-source projects or overseas developers is slow and expensive. Expect the first big court case to set the rules.

Security holes get found faster

This cuts both ways. Defenders can audit closed-source software they depend on. Attackers can do the same, looking for flaws in apps that were “secure” partly because nobody had bothered to read them. We covered the defensive side of that in how AI is finding software vulnerabilities.

Can you stop AI from reverse engineering your app?

Not completely. Developer Timo Zhang built Blind LLM Agents, a tool that hardens shipped code with renaming, string encryption, build-to-build variation, decoy code and integrity checks. His argument is that “classic obfuscation was built for humans and static tools, not LLMs that reason about meaning.” But he’s upfront about the limits: “We raise the cost — we don’t claim it is unbreakable.”

Intezer reached a similar view. Encryptors and obfuscators “raise the bar” for AI agents, but they don’t close the door. For most companies, the practical defenses look like this:

  1. Keep the valuable logic on the server. Code that never ships to the user’s device can’t be decompiled from it. Pricing rules, fraud checks and proprietary models belong behind an API.
  2. Assume your client code will be read. Never embed secrets, keys or admin backdoors in an app. Our story on leaked API keys in AI coding plugins shows how often this goes wrong.
  3. Use obfuscation as friction, not protection. It slows casual copying and automated tools. It won’t stop a determined, funded team.
  4. Compete on what can’t be copied. Data, network effects, service quality and speed of shipping matter more when features are easy to clone.

What this means for you

If you don’t write software, you’ll feel this mainly through security updates. Cheaper reverse engineering means bugs in the apps you use get found faster, by good and bad actors. The gap between a flaw being found and being exploited is shrinking, so the old advice matters more: turn on automatic updates for your phone, browser and apps, and don’t sit on “update available” notices for weeks.

It also means more lookalike apps. Some will be legitimate open-source alternatives. Others will be fake versions carrying malware, built by copying a real app’s look and behavior. Download software only from official stores or the maker’s own site, and be wary of “free” versions of paid tools. Our guide to spotting phishing scams covers the warning signs.

Our take: AI reverse engineering is a net positive for security research and a real threat to software businesses that rely on secrecy. Companies that treat their client code as public, and protect what really matters on their servers, will be fine. Companies whose whole moat is “nobody can read our app” have a problem.

Frequently asked questions

Is reverse engineering software legal?

It depends on the country, the license and the purpose. In the US, reverse engineering for interoperability or security research has some legal protection, but many software licenses forbid it and copying protected code into a competing product can infringe copyright. If you’re doing it commercially, get legal advice first.

Can AI decompile any app?

Not perfectly. Intezer’s Finder rebuild still had gaps around features that depend on other system services, and heavily protected software is harder. But for many apps, AI can now produce working code good enough to study, modify or extend.

Does AI reverse engineering help hackers?

Yes, and it helps defenders too. Attackers can find flaws in closed-source software faster. Security teams can rebuild malware to study it and write detection rules. Which side benefits more depends on who adopts the tools faster.

What is code obfuscation?

Code obfuscation is the practice of deliberately making compiled code harder to read without changing what it does, for example by scrambling names and encrypting text. It adds friction for reverse engineers, human or AI, but it can’t make code impossible to analyze.

Should I worry about apps I use being cloned?

Not directly, but watch for fake copies. Cloned apps carrying malware are an old trick that’s getting cheaper. Stick to official app stores and developer websites, and check reviews and the publisher name before installing.