Loading story…
A defect in Lean’s low-level String.Pos.Raw.extract function affected stable releases through Lean 4.33.1, creating a semantic mismatch between its logical evaluator and compiled native code. At an astronomically large byte offset, Lean’s logical definition returned an empty string for a one-byte slice, while native code returned the original string; researchers used the discrepancy to construct a contradiction and a proof-of-concept false proof of Fermat’s Last Theorem.
The issue is not a flaw in Lean’s proof-checking kernel. Exploitation requires native_decide, which trusts native evaluation and expands the trusted computing base to include the compiler and runtime behavior. A memory-safety fix was opened roughly 90 minutes after disclosure, followed by a correction for the semantic mismatch; the fixes are included in Lean v4.34.0-rc1.
See affected versions and whether adversaries are exploiting it.
4 events from the most recent confirmed update back to the earliest known activity.
Five days after the report, contributor Rob23oba fixed the remaining mismatch between Lean's logical evaluator and compiled native behavior, closing the issue. The resulting patch was incorporated in Lean v4.34.0-rc1.
The memory-safety fix opened by hargoniX was merged roughly three hours after it was filed.
About 90 minutes after the issue was reported, contributor hargoniX opened a fix for the memory-safety issue in the low-level string extraction behavior.
Researchers identified a semantic mismatch in Lean's String.Pos.Raw.extract and used native_decide to derive a contradiction equating an empty string with a non-empty string. They used the contradiction in a proof-of-concept purported proof of Fermat's Last Theorem affecting stable Lean releases through 4.33.1.
See whether adversaries are exploiting this yet, and where the affected versions run in your environment.
4 references tracked. Mallory keeps watching after this page renders.
malware.news
Open sourceblog.trailofbits.com
Open sourcegithub.com
Open sourceanthropic.com
Open sourceMap indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Trail of Bits found a high-severity underconstrained-input vulnerability in Miden’s zero-knowledge virtual machine core cryptographic library that could let a malicious prover forge Falcon signatures and steal funds from accounts using Falcon key pairs. The flaw in the MASM `mod_12289` procedure allowed a prover-supplied remainder to reach a 32-bit subtraction without first being validated as a 64-bit value. Auditors developed AI-assisted tooling for the MASM-based stack, including a language server, decompiler, static-analysis engine, linter, and Lean formal model of the VM executor. The static analysis identified more than 400 publicly reachable instances of insufficient type validation, while formal verification generated 95 machine-checked proofs and found additional defects in `rotr` and `wrapping_mul`; Miden has adopted the static-analysis engine for future core-library changes.
Sep 18, 2026
Researchers disclosed **CVE-2026-31431**, dubbed **CopyFail**, a Linux local privilege escalation flaw in the kernel crypto subsystem's `algif_aead`/`AF_ALG` AEAD path that lets a low-privileged local user corrupt the page cache of readable files and gain root. The bug traces to a 2017 in-place AEAD optimization introduced in Linux 4.14 by commit `72548b093ee3`, which allowed splice-backed file pages to be reused as writable crypto buffers; attackers can then achieve a controlled write into the cached copy of targets such as setuid binaries while leaving the on-disk file unchanged. Researchers published a compact Python proof of concept said to work reliably across major distributions including Ubuntu, RHEL, SUSE, Amazon Linux, Debian, Fedora, and Arch, and warned the shared page cache also makes the flaw relevant for container escapes, CI runners, and multi-tenant cloud hosts. Upstream fixed the issue by reverting `algif_aead` to out-of-place operation in Linux **7.0**, **6.19.12**, and **6.18.22**, while older long-term branches initially lagged because the patch did not apply cleanly and required backports or workaround patches. Major vendors and downstream projects began shipping advisories and kernel updates, and temporary mitigations included disabling or blacklisting `algif_aead`, blocking `AF_ALG` socket creation with seccomp or eBPF, or using the kernel boot parameter `initcall_blacklist=algif_aead_init` where the feature was built in. After public exploit code spread and additional exploit ports appeared, **CISA added the flaw to its Known Exploited Vulnerabilities catalog**, and defenders published detections focused on suspicious `AF_ALG`, `splice()`, and setuid execution activity because traditional file-integrity checks may miss the in-memory-only tampering.
Aug 25, 2026
Researchers from Germany’s Helmholtz Center for Information Security disclosed **LoongLeak**, a cache leakage vulnerability affecting Loongson processors built on the **LoongArch** ISA. The flaw is tied to an instruction documented as leaving 32 bits of a memory register in an “uncertain” state, which the researchers found can leak data from the shared **L1 data cache** across applications and the operating system. Their testing showed attackers could extract sensitive information including full-disk **AES** keys, partial root password hashes, and data useful for defeating **ASLR** and stack canaries within seconds. The researchers said exploitation is possible from unprivileged user space, containers, and guest virtual machines, enabling leakage across VM boundaries into host data. Software-only mitigations were described as insufficient, though Loongson reportedly addressed the issue in an update to its **3A6000** processor. A cache-eviction mitigation was also noted, with an estimated performance impact of up to **1.4%**.
Aug 14, 2026