Anthropic’s Claude Fable 5 has generated a functional, bootable NT-compatible Windows kernel written in Rust from an empty directory in just 38 minutes of active model work.
Documented by security researcher Matt Suiche and Tolmo’s threat research agent “Twinkle” on June 22, 2026, the project, called ntoskrnl-rs, represents one of the most consequential demonstrations of AI-authored systems software to date.
Anthropic’s Fable 5 handled the entire core scaffolding in a single contiguous session, generating approximately 5,100 lines of code across 27 files.
The output spanned every critical subsystem: scheduler, memory manager, trap and interrupt machinery, object manager, and I/O manager, organized precisely to mirror ntoskrnl’s own subsystem layout.
Fable 5 Writes Windows Kernel in Rust
The kernel booted successfully in the QEMU emulator and passed all 14 in-kernel self-tests, exiting with the project’s standing pass contract of exit code 33.
While the full wall-clock session ran roughly four and a half hours, the human operator was away from the keyboard for most of it. The model’s actual active work window was 38 minutes.
Key technical metrics from the session include 197 assistant turns, 110 tool calls (45 writes, 25 Bash, 18 edits), 43 files touched, approximately 407K output tokens on 11K fresh input, and a time-to-bootable-core of 38 minutes.
What separates this from ordinary code generation is the model’s demonstrated capacity for unsupervised, low-level systems reasoning. Fable 5 caught and corrected two critical bugs mid-generation without any human intervention.
The first was an EOI ordering bug: the model found that the end-of-interrupt signal must be issued before any potential context switch, since preempting mid-dispatch would deadlock the local interrupt controller.
The second was an IRQL emulation bug: when host tests returned 11/12, Fable diagnosed that the interrupt request level emulation relied on a single global atomic shared across test threads, corrected it to a per-thread thread_local variable mirroring real per-CPU hardware behavior, and passed 12/12.
The model also embedded architectural commentary in the code, explaining why the NT GDT selector ordering aligns with the IA32_STAR MSR format forward-looking ABI, which accounts for future syscall support without a redesign.
Fable 5 produced roughly 40% of the project’s from-scratch code in just 3% of total turns. The remaining 97% of turns eight days of iterative, debug-heavy bring-up ran on Claude Opus 4.8, which extended the kernel to load unmodified Windows kernel drivers and execute real Windows binaries, including sort.exe, choice.exe, and cmd.exe.
The handoff was deliberate. Fable 5 carries aggressive cybersecurity safety classifiers broad enough to flag adjacent defensive work.
Fable launched publicly on June 10, 2026, as the limited release of Anthropic’s Mythos cybersecurity model tier, and within days, a U.S. government export-control directive forced Anthropic to suspend access entirely.
Fable 5 identified this gap itself, unprompted, flagging the dispatcher lock hand-off, spinlocks, and DPC queue as the highest-risk paths, and recommending loom for exhaustive concurrency exploration and Miri for undefined behavior detection.
This is the core security implication: authoring capability has outpaced verification. A model can produce the Trusted Computing Base of an x86_64 kernel faster than any human team can audit it.
Until formal verification, property testing, and concurrency model checkers close that gap, an AI-authored kernel remains a booting artifact of unknown correctness, and unknown correctness has no place in a TCB.
Tolmo stated that the internet’s critical infrastructure runs on aging C codebases, maintained largely because rewriting a TCB has historically been too costly and too risky.
An AI-authored Rust kernel represents a double lever: Rust eliminates the memory-safety bug classes that dominate OS CVEs, while AI eliminates the human-cost bottleneck of the rewrite itself.
Once verification tooling matures, the economic case for leaving legacy C in place collapses, and large portions of the stack become candidates for AI-driven, memory-safe rewrites.
Follow us on Google News , LinkedIn and X to Get More Instant Updates. Set Cyberpress as a Preferred Source in Google.
