AI Tool Identifies Vulnerabilities in Decades-Old Apple II Program

Microsoft Azure CTO Mark Russinovich demonstrated that AI can now analyze decades-old machine code to uncover vulnerabilities, using his own Apple II program from 1986 as an example. The program, named Enhancer, extended Applesoft BASIC by allowing GOTO, GOSUB, and RESTORE commands to use variables or expressions instead of line numbers. Using Claude Opus 4.6, the AI decompiled the 6502 machine language and detected several issues, including a subtle bug where the program would point to the wrong line if a destination was missing instead of generating an error. Although these vulnerabilities have little practical impact today, the demonstration highlights AI’s growing ability to audit and reverse-engineer embedded and legacy code, which is significant because billions of older microcontrollers still operate worldwide with potentially fragile or unaudited firmware. Experts suggest this capability could benefit both cybersecurity defenders and attackers in identifying weaknesses in old or poorly maintained code. While the example involves Apple II software, the broader implications emphasize the importance of AI in modern security research and automated vulnerability detection.