Public preview · Registration and live billing are not open.
Anti-AIClonePILOT

Log date · Thu

Binary Ninja 6.0 Krypton: a first-party MCP server and binary similarity

Binary Ninja 6.0 shipped a first-party MCP server for LLM clients, faster analysis, and binary similarity, making LLM-driven analysis a mainstream workflow.

What happened on this date

On 2026-09-03, Vector 35 announced Binary Ninja 6.0, codenamed Krypton; the stable/6.0.10601 release page in the binaryninja-api repository is timestamped 2026-09-03T23:34Z (UTC). Highlights listed in the post include analysis up to 2.16x faster than 5.3 in their benchmarks with up to 55% lower peak memory; a first-party MCP server that exposes triage, program structure, disassembly, pseudo-C, ILs, and renaming and typing operations to LLM clients, with a GUI variant in every edition including Free and a headless variant in Commercial and Ultimate on macOS and Linux; Objective-C support for method type strings in iOS 27 shared caches and renaming of objc_msgSend stubs; and binary similarity in Ultimate with BinDiff and WARP providers. The post did not mention obfuscation-specific features.

Why it matters for protecting app logic

When a mainstream disassembler ships an MCP interface even in its free edition, letting an LLM drive a reversing tool stops being a research prototype and becomes an ordinary workflow. Your threat model should therefore assume an analyst can hand your binary to an agent that repeatedly queries, renames and assigns types. Binary similarity and faster analysis make version-to-version comparison cheaper, so what changed in each update is easier to locate. Objective-C metadata remains one of the most informative parts of an iOS app. The boundary is that the post gives no data on how these capabilities perform against transformed code, and we should not guess how they fare on hardened binaries.

Checks you can run

First, inventory the Objective-C selectors, class names, Swift metadata and exported symbols that remain in your release build, and decide which are required at runtime and which can be removed. Second, assume each update will be diffed against the previous one, and move changes to sensitive rules to the server where possible. Third, when assessing your own app internally, include LLM-assisted analysis as a scenario in the threat model, using tools you are licensed for and your own builds. Fourth, when reviewing a protection vendor's claims, check whether the tool versions they tested against already include capabilities like these.

Your checklist

  • Inventory ObjC/Swift metadata and exported symbols in your release build
  • Assume every update is diffed; move sensitive changes server-side
  • Check whether protection assessments used current analysis tool versions

Primary sources

All sources and reproduction status ↗