The question behind the finding
What would it take to find a memory-safety bug in the part of Rust that is supposed to rule them out? Rust's promise is a bargain: write 'safe' code and the compiler guarantees whole classes of memory error cannot happen. That guarantee leans on the standard libraries underneath, which contain carefully reviewed unsafe code. If a library gets it wrong, a program that never types the word unsafe can still misbehave.
Microsoft's FORGE Lab says that is what it found. In a Microsoft Security Blog post dated November 9, the lab reports memory-safety bugs in Rust's core and std libraries, including cases where safe Rust code can trigger undefined behaviour. The post's title credits agentic tools, and its summary names one, Hyenas, described as a FORGE research system.
Here is what that does not tell us, at least from the title and summary the Journal has read. It doesn't say how many bugs, how severe, whether they are fixed, or which Rust versions are affected. It doesn't say how much of the finding was the agent and how much was the researchers steering it, which is the difference between a tool result and a method. And it is the lab's own account; the Rust project's response is the independent check. One more thing, because it gets lost: finding soundness holes in a standard library is evidence the review process works, not that the language's bargain is void.
What I'm watching for
Two things would firm this up: the upstream issue or advisory for each bug, with the fix version, and a count of what the agent flagged that turned out to be wrong. A bug-finder's false-positive rate is half its result.
