A newly released AI model, Claude Fable 5, has made a significant advancement in autonomous systems programming by generating a bootable Windows NT-style kernel in Rust in just 38 minutes.
The project, titled ntoskrnl-rs, began as an empty repository and evolved into a functioning x86_64 kernel that boots in QEMU and passes all internal self-tests. This raises important questions about the future of developing and verifying trusted computing bases (TCB).
The kernel produced by Fable 5 includes core subsystems that are typically complex and security-critical, such as the scheduler, memory manager, interrupt and trap handling mechanisms, object manager, and I/O manager, as reported by Tolmo.
Together, these components form the TCB, where any flaw can undermine the integrity of the entire system. Notably, the AI generated approximately 5,100 lines of Rust code across 27 files and structured the kernel closely in alignment with Microsoft’s ntoskrnlarchitectural design.
During execution, the model autonomously planned and executed subsystem dependencies, initialized low-level constructs such as the Global Descriptor Table (GDT) and Interrupt Descriptor Table (IDT), and mapped hardware-level abstractions, such as Interrupt Request Level (IRQL), to CR8 registers.
It successfully booted the kernel in QEMU, yielding a clean test output with all validation checks passing, including memory allocation, thread scheduling, synchronization primitives, and basic I/O operations via a null driver interface.
One of the most remarkable aspects of this experiment was the model’s ability to identify and fix its own bugs in real time. For example, it detected a potential deadlock risk in interrupt handling due to delayed End-of-Interrupt (EOI) signaling and corrected the execution order.
It also resolved an IRQL misimplementation by replacing a global atomic variable with thread-local storage to emulate per-CPU behavior better. These corrections suggest reasoning beyond simple pattern generation, indicating an understanding of kernel-level concurrency and hardware interactions.
Despite the rapid development, the kernel remains minimal. Initially, it lacked user-mode support and could not execute external programs. However, subsequent enhancements extended its capabilities, allowing it to load unmodified Windows drivers and run select user-mode binaries, such as sort.exe and cmd.exe, via a custom PE loader and API shims.
This progress underscores the potential for AI-generated kernels to serve as controlled environments for malware analysis, driver testing, and syscall tracing.
The experiment also highlights a significant gap between code generation and verification. While the kernel compiles and runs, its correctness and security have not been confirmed.
The model itself proposed advanced verification techniques, including concurrency testing using Loom and undefined behavior detection with Miri. This reflects an emerging reality in cybersecurity: AI can now produce complex, low-level systems faster than humans can audit them.
Below is a summary of the Fable 5 kernel generation metrics:
| Metric | Value |
|---|---|
| Invocations | Single continuous run |
| Assistant Turns | 197 (including 110 tool calls) |
| Files Modified | 43 files across 63 operations |
| Code Generated | ~5,100 lines across 27 files |
| Execution Time | 38 minutes (core build) |
| Token Usage | ~407K output tokens |
| Self-Test Results | All tests passed (exit code 33) |
The broader implications for cybersecurity are profound. AI-driven development could accelerate the shift from legacy C-based infrastructure to memory-safe Rust implementations.
However, without robust verification frameworks, AI-generated kernels remain experimental rather than production-ready systems. As both threat actors and defenders gain access to such capabilities, the focus will shift to validating, auditing, and controlling AI-generated code at the lowest levels of computing.
This milestone marks a turning point: AI is no longer just generating application-layer code but is actively influencing the foundations of operating systems, redefining both opportunities and risks in modern cybersecurity.
Follow us on Google News, LinkedIn, and X to Get Instant Updates and Set GBH as a Preferred Source in Google.
Attackers abused compromised ConnectWise Cloud Backup credentials to access Microsoft 365 customer environments, prompting ConnectWise…
Anthropic says an internal Claude research model reconstructed how the National Security Agency generated seeds…
Two WordPress Core vulnerabilities can be chained to execute remote code without authentication, according to…
A newly disclosed OpenBSD kernel exploitation chain combines a shared memory accounting flaw with a…
GhostAction attackers have expanded their GitHub credential theft campaign with a second stage that disguises…
Indian authorities have warned that SideCopy hackers are targeting Indian organizations, including academic institutions, with…