Intezer uses AI agents to rebuild Slack, Apple Finder and a TinyTurla-NG (Turla) backdoor from binaries, lowering the bar for reverse engineering
Matéria consolidada. Reúne 2 matérias que o Tifon tinha publicado separadamente sobre a mesma história. A IA reescreveu o relatório com o que as 2 fontes trouxeram em 11/11/2026 00:06; os links antigos trazem para cá.
Intezer researchers used AI agents to rebuild working source code for Slack, Apple's Finder and a TinyTurla-NG sample, a backdoor family that Cisco Talos links to the Turla group (helpnetsecurity.com). The agents worked only from shipped programs and vendor documentation. Intezer senior security researcher Nicole Fishbein says AI is lowering the bar, so work that once needed seasoned experts is now within reach of small teams and individuals (darkreading.com). This is a proof of concept, not an intrusion. Neither source reports a vulnerability, a CVE, a patch or exploitation in the wild (darkreading.com, helpnetsecurity.com). The risk is that vendors relying on compiled binaries to protect source code and IP lose that barrier, and that the same method speeds up malware analysis (darkreading.com, helpnetsecurity.com). The Dark Reading account says it could not locate the original Intezer write-up, so details rest on the two news reports. Help Net Security (published 2026-11-10, after Dark Reading) is the more detailed source and rates the criticality as medium.
Detalhes técnicos
- Tooling: Codex, running GPT-5.6-sol-high, orchestrated the workflow and made the engineering decisions, while Kimi K3 filled three separate roles. The agents produced about 3,100 lines of source in two files (helpnetsecurity.com).
- Success criteria: The team set five criteria. The code had to be readable enough to maintain, clean and idiomatic, compiled only from what the agents wrote, free of bundled third-party libraries, and identical in behavior to the original (helpnetsecurity.com).
- Slack: The rebuilt client ran well enough that the team added an AI text autocomplete feature powered by a small local model (helpnetsecurity.com). Slack's JavaScript source offered little the researchers could not easily unpack (darkreading.com). Claude analyzed Slack's minified client to identify the third-party libraries it contains, among more than 500,000 functions. Because Slack's server is not downloadable, the agents built a Slack-compatible server on SQLite, guided by network inspection of the real app and Slack's documentation (helpnetsecurity.com).
- Finder: The rebuild is a 25.8 MB executable. It still has gaps around online-only files, sync status and actions that depend on other macOS services (helpnetsecurity.com). Finder was harder than Slack because much of its behavior depends on files handled by cloud services and on other macOS services (darkreading.com).
- TinyTurla-NG: The rebuilt DLL is 296,960 bytes against 291,840 for the original, with 1,072 functions against 1,041 and identical imported symbols. BinDiff matched 1,034 original functions and scored the pair at 92.8% similarity with 98.7% confidence. The researchers say this measures structural shape only (helpnetsecurity.com). The resulting replicant reproduces the malware's full working pipeline, including how it receives and executes commands and its file transfer mechanisms (darkreading.com).
- Behavior validation: A separate harness compares each rebuilt function with its original using Binary Ninja's high-level intermediate language (helpnetsecurity.com).
- Not disclosed: The article does not say how long the rebuilds took or what they cost (helpnetsecurity.com).
Remediação
- No patch or configuration change is involved, and neither source reports a vendor response from Slack, Apple or Cisco (helpnetsecurity.com).
- Intezer advises vendors to use encryptors and obfuscators, which the researchers say raise the bar for AI agents. The article does not report testing how much these tools slow an agent down (helpnetsecurity.com).
- Defenders should assume that AI-assisted reverse engineering of malware and proprietary software is now cheaper and faster, with implications for cyber defense and IP theft (darkreading.com).
Relatos divergentes
- Hardest target: Dark Reading says the malware was the hardest target, but its wording on the Apple and Salesforce apps, which it says proved the most difficult to wrangle, is garbled and unclear (darkreading.com). Help Net Security rates TinyTurla as the hardest of the three (helpnetsecurity.com). The Help Net Security account is the more recent and more detailed.
- Malware naming: Dark Reading refers to the target as Turla malware (darkreading.com). Help Net Security specifies a TinyTurla-NG sample, a backdoor family that Cisco Talos links to Turla (helpnetsecurity.com).
Linha do tempo
-
Intezer uses AI agents to rebuild Slack, Apple Finder and a TinyTurla-NG backdoor from binaries
ler a análise desta fonte ocultar
Intezer researchers used AI agents to rebuild working source code for Slack, Apple's Finder and a TinyTurla-NG sample, a backdoor family that Cisco Talos links to the Turla group, working only from shipped programs and vendor documentation. The rebuilt Slack client ran well enough that the team added an AI text autocomplete feature powered by a small local model. The work is a proof of concept about reverse engineering, not an active intrusion: the article reports no vulnerability, no CVE and no attack in the wild. The risk is that vendors relying on compiled binaries to protect their source code lose that barrier, and the same method can be applied to malware. Criticity is medium because there is no direct exposure, only a shift in how hard it is to recover code.
Technical Details
- Codex, running GPT-5.6-sol-high, orchestrated the workflow and made the engineering decisions, while Kimi K3 filled three separate roles. The agents produced about 3,100 lines of source in two files.
- The team set five success criteria: readable enough to maintain, clean and idiomatic, compiled only from what the agents wrote, free of bundled third-party libraries, and identical in behavior to the original. TinyTurla was rated the hardest of the three targets.
- For the TinyTurla-NG DLL, the rebuild is 296,960 bytes against 291,840 for the original, with 1,072 functions against 1,041 and identical imported symbols. BinDiff matched 1,034 original functions and scored the pair at 92.8% similarity with 98.7% confidence. The researchers say this measures structural shape only.
- Behavior was checked separately with a harness that compares each rebuilt function with its original using Binary Ninja's high-level intermediate language.
- Claude analyzed Slack's minified client to identify the third-party libraries it contains, among more than 500,000 functions. Because Slack's server is not downloadable, the agents built a Slack-compatible server on SQLite, guided by network inspection of the real app and Slack's documentation.
- The Finder rebuild, a 25.8 MB executable, still has gaps around online-only files, sync status and actions that depend on other macOS services.
- The article does not say how long the rebuilds took or what they cost.
Remediation
- Intezer advises vendors to use encryptors and obfuscators, which the researchers say "raise the bar" for AI agents. The article does not report testing how much these tools slow an agent down.
- The article reports no vendor response from Slack, Apple or Cisco, and no patch or configuration change is involved.
-
darkreading.com ↗ · pub. 10/11/2026 10:00 · coletado 10/11 10:00 primeiro registro só resumo do feed
Intezer Shows AI Lowers the Bar for Reverse Engineering Slack, Apple Finder and Turla Malware
ler a análise desta fonte ocultar
Intezer researchers used AI tools to reverse-engineer three targets: Slack, Apple's Finder app and the Turla malware (darkreading.com). Their finding is that work which once demanded seasoned experts is now within reach of even small teams and individuals (darkreading.com), and Intezer senior security researcher Nicole Fishbein describes the effect as AI lowering the bar (darkreading.com). The malware was the hardest target, while rebuilding Apple's and Salesforce's proprietary apps from the ground up proved the most difficult to wrangle only moderately (darkreading.com). Researchers warn the shift has implications for cyber defense, intellectual property (IP) theft, and beyond (darkreading.com). No CVE, patch or in-the-wild exploitation is reported; the risk lies in faster analysis of malware and proprietary software, not in a flaw in any product. The Intezer write-up of the experiment could not be located during research, so the account rests on the Dark Reading report.
Technical Details
- Slack was the easier target: its JavaScript source code offered little that the researchers couldn't easily unpack (darkreading.com).
- Finder was harder because much of its behavior depended on files handled by cloud services and on actions that relied on other macOS services (darkreading.com).
- The Turla work ended with a replicant that reproduces its full working pipeline: how it receives and executes commands, its file transfer mechanisms (darkreading.com).