Built with WordPress and OceanWP, hosted on Cloudflare. Domain registered in 2025.
xsourcesec.com is built with WordPress and the OceanWP theme, hosted on Cloudflare. It measures traffic with Cloudflare Web Analytics. It sends email through Amazon SES.
| Technology | Version | First seen |
|---|---|---|
| CMS & Site Builders1 | ||
| – | 7 Nov 2026 | |
| WordPress Themes1 | ||
| – | 7 Nov 2026 | |
| WordPress Plugins3 | ||
| – | 7 Nov 2026 | |
| Web Servers, Languages & Backend Frameworks1 | ||
| – | 7 Nov 2026 | |
| CDN & Hosting1 | ||
| – | 7 Nov 2026 | |
| Analytics & Tag Management1 | ||
| – | 7 Nov 2026 | |
| Transactional Email2 | ||
| – | 7 Nov 2026 | |
| Security, WAF & Bot Protection1 | ||
| – | 7 Nov 2026 | |
Registration from the registry's RDAP record, the certificate the site serves, and its rank in public traffic lists.
The crew hasn’t read this site’s colors yet. We read a palette when we capture a site’s screenshot, busiest sites first. Meanwhile, browse palettes from other sites.
Counted from every stack change we've recorded: of the sites we watched running one of this site's tools, how many went on to add another. A pattern across sites, not a forecast for this one.
From what the site says about itself, and account IDs it shares with other sites.
mail.xsourcesec.comnelly.ns.cloudflare.comgabriel.ns.cloudflare.comv=spf1 a mx ip4:108.167.161.41 include:websitewelcome.com ~allgoogle-site-verification=hgoEGaTCNXBddmoLf0EEyE9XoM6nM3cs1QWuQAQ6Wh8v=DMARC1; p=none; rua=mailto:20f91893f9f84d94b8d456d9e9f8dd38@dmarc-reports.cloudflare.net,mailto:eva@xsourcesec.com# Content signals: https://contentsignals.org/ User-agent: * Content-Signal: search=yes, ai-input=yes Disallow: /wp-admin/ Allow: /wp-admin/admin-ajax.php Sitemap: https://xsourcesec.com/sitemap_index.xml
Generated by Rank Math SEO, this is an llms.txt file designed to help LLMs better understand and index this website. # XSource Security > XSource Security provides continuous AI security testing for companies that ship AI: ongoing AI red teaming and AI penetration testing for chatbots, AI agents and LLM applications, plus one-off AI Security Assessments. It is run by Eva Crystal (credited as 0xiviel), an independent security researcher and penetration tester in Christchurch, New Zealand, with 10 published CVEs, a co-credited liboqs CVE and six merged upstream fixes, including in the Linux kernel. Its source-level audits of AI infrastructure, firmware and cryptography are published under Research. ## Sitemaps - [XML Sitemap](https://xsourcesec.com/sitemap_index.xml): Includes all crawlable and indexable pages. ## Posts - [Shorter than a signature: an unsigned underflow in CROSS](https://xsourcesec.com/shorter-than-a-signature-unsigned-underflow-in-cross/): A signed message shorter than one signature made the CROSS reference implementation compute a length close to 2^64 and pass it to memcpy, before verification ran. CVE-2026-41509, scored 9.8 Critical. - [A ceiling on the wrong key: uncapped video frames in vLLM](https://xsourcesec.com/a-ceiling-on-the-wrong-key-uncapped-video-frames-in-vllm/): Seventy-four bytes of JSON took a vLLM API server's peak memory from 2,271 MiB to 13,629 MiB, and the frame-count ceiling that should have stopped it was sitting on a key the Qwen video samplers never read. - [When null is a user: private memory disclosure in AnythingLLM](https://xsourcesec.com/when-null-is-a-user-private-memory-disclosure-in-anythingllm/): An anonymous embed visitor received the instance owner's private Memories. Not prompt injection: a missing identity that collapsed into the owner's own storage scope. GHSA-8wh8-3v68-whwj, fixed under eleven hours after the report. - [Inside the Lenovo SMM escalation chain](https://xsourcesec.com/inside-the-lenovo-smm-escalation-chain/): Part 1 of 4. The Lenovo WmiBiosSetupSmm disclosure (CVE-2026-10587 to -10590), coordinated with Lenovo PSIRT under case LEN-213632 and published in advisory LEN-220440 on 14 July 2026. - [Validating the unvalidated: SMI input handling](https://xsourcesec.com/validating-the-unvalidated-smi-input-handling/): Part 2 of 4: the Lenovo WmiBiosSetupSmm disclosure. Start with Part 1 for the architecture and the escalation chain. - [Authorization boundaries in SMM communication](https://xsourcesec.com/authorization-boundaries-in-smm-communication/): Missing size and buffer checks make a handler fragile. Two more gaps make it usable: it authenticates nobody, and it hands the attacker a map. This part is about the authorization boundary in SMM communication, where it sits, how it's crossed here, and what one well-built module in the same firmware tells us about doing it right. - [Coordinated disclosure with Lenovo PSIRT](https://xsourcesec.com/coordinated-disclosure-with-lenovo-psirt/): Firmware fixes are slow by nature: long validation cycles, multiple vendors (Lenovo and Insyde here), and devices that live for years. A 90-day clock with real status updates gave the fix time to ship before any technical detail went public, which is the entire point: users were protected first, and the write-ups you're reading now add nothing an attacker didn't already have to work for. My thanks to the Lenovo PSIRT team for being responsive and precise throughout. ## Pages - [AI Security Assessment](https://xsourcesec.com/ai-security-assessment/): A full AI red teaming and AI penetration test of one AI application, delivered as a written report with reproduced findings and remediation. Typical engagement window: 1 to 2 weeks. - [Continuous AI Security](https://xsourcesec.com/continuous-ai-security/): AI red teaming and AI penetration testing that runs alongside your releases: one AI application, scheduled test rounds each month, and the changes you flag in time tested within the agreed scope before they reach your users. - [Research](https://xsourcesec.com/research/): Published CVEs, advisories and upstream fixes in AI infrastructure, firmware and cryptography, with the source-level audit lines and public audit reports. - [Security](https://xsourcesec.com/security/): How to report a security issue in this site or in anything XSource Security publishes, what happens next, and the safe harbour that covers good-faith research. - [Cryptography & PQC Audit](https://xsourcesec.com/services/cryptography-pqc/): Review of cryptographic implementations and the code around them: post-quantum and classical schemes, TLS stacks, signature verification and key handling. The area behind most of the published CVEs. Typical engagement window: 3 to 4 weeks. - [Firmware & Accelerator Audit](https://xsourcesec.com/services/firmware-accelerator/): Implementation review of boot-chain firmware, privileged handlers and the drivers that sit between an AI workload and the hardware. Typical engagement window: 3 to 4 weeks. - [AI Infrastructure Audit](https://xsourcesec.com/services/ai-infrastructure/): Source review and live testing of the layer your AI product runs on: inference servers, LLM gateways, application platforms, RAG pipelines, agents and MCP integrations, accelerator runtimes. Typical engagement window: 2 to 4 weeks. - [Source-level audits](https://xsourcesec.com/services/): One-off source-level security audits of AI infrastructure, firmware and cryptography, with a written scope and a fixed price before any work starts. Every finding is reproduced against the build in scope and written up at advisory grade: the same standard the published CVEs had to meet. - [Privacy](https://xsourcesec.com/privacy/): What this site collects, why, where it is held and how to ask about it. XSource Security is a sole-trader business in Christchurch, New Zealand, and handles personal information under the New Zealand Privacy Act 2020. - [Eva Crystal](https://xsourcesec.com/eva-crystal/): Independent security researcher and penetration tester, founder of XSource Security. Christchurch, New Zealand. - [Blog](https://xsourcesec.com/blog/) - [Home](https://xsourcesec.com/) ## Disclosures - [liboqs LMS verification accepts signatures from another parameter set](https://xsourcesec.com/disclosures/ghsa-xv6c-h66p-6mfq/): LMS verification did not bind the signature and public key to the parameter set of the object in use. Reported publicly in issue #2568 and credited as an independent report in the joint advisory; CVE-2026-104255 assigned. - [vLLM Qwen2-VL / Qwen3-VL video sampler memory exhaustion](https://xsourcesec.com/disclosures/cve-2026-105758/): The Qwen2-VL and Qwen3-VL video samplers bound on the request-controlled max_frames value, which the server-side frame ceiling does not reach. One unauthenticated request to /tokenize exhausts API-server memory before admission control. - [AnythingLLM: owner's private Memories disclosed to anonymous embed-widget visitors](https://xsourcesec.com/disclosures/ghsa-8wh8-3v68-whwj/): The embed handler passed a visitor-supplied username where a user object was expected, so the memory lookup collapsed to the single-user scope and the instance owner's private Memories were appended to the system prompt on an unauthenticated route. - [Missing authentication - WMI-triggered SMI handler](https://xsourcesec.com/disclosures/cve-2026-10590/): Missing authentication allows WMI commands to trigger SMI handlers arbitrarily, reaching SMM (ring -2) code paths. - [SMM protected-memory address disclosure](https://xsourcesec.com/disclosures/cve-2026-10588/): Discloses the address of protected SMM memory to a local privileged attacker, weakening SMRAM protections. Ring -2. - [SMM out-of-bounds write - power-management settings](https://xsourcesec.com/disclosures/cve-2026-10587/): Out-of-bounds write in System Management Mode lets a local privileged attacker modify power-management settings. Ring -2, below Secure Boot. - [SMM out-of-bounds write - code execution](https://xsourcesec.com/disclosures/cve-2026-10589/): Out-of-bounds write in System Management Mode enabling code execution at SMM (ring -2) by a local privileged attacker. - [Mbed TLS ECDH calc_secret buffer overflow](https://xsourcesec.com/disclosures/cve-2026-35336/): Possible buffer overflow in mbedtls_ecdh_calc_secret() when the output buffer is too small - up to remote code execution. - [Eight cases submitted in February 2026 covering AMD Xilinx Bootgen, U-Boot, EDK2 and coreboot](https://xsourcesec.com/disclosures/8-cases-submitted-covering-amd-xilinx-bootgen-u-boot-edk2-and-coreboot/): Six were closed in August 2026 on scope, after ZDI suspended acquisition of embedded-systems vulnerabilities six weeks after the submissions; a re-review has been requested. Two remain under investigation (AMD Xilinx Bootgen, U-Boot). - [CROSS post-quantum signature buffer overflow](https://xsourcesec.com/disclosures/cve-2026-41509/): Integer underflow of mlen in crypto_sign_open() in the CROSS reference implementation leads to a heap buffer overflow during signature verification. - [oqs-provider NULL-check hardening](https://xsourcesec.com/disclosures/oqs-provider-pr-767/): Duplicate NULL-check correction in oqsx_get_hybrid_params(). Fix contributed as upstream PR #767, merged and credited. - [oqs-provider double-free in signature handling](https://xsourcesec.com/disclosures/oqs-provider-pr-766/): Double-free vulnerability in oqs-provider. Fix contributed as upstream PR #766, merged and credited. - [Mbed TLS out-of-bounds read](https://xsourcesec.com/disclosures/cve-2026-34876/): Out-of-bounds read in mbedtls_ccm_finish(): an oversized tag_len in the multipart CCM API exposes adjacent CCM context data. Discovered through the Intigriti Trusted Firmware program. - [Mbed TLS serialized session/context memory safety](https://xsourcesec.com/disclosures/cve-2026-34877/): Insufficient protection of serialized SSL session/context data can lead to memory-safety issues when restoring tampered or corrupted state. Credited jointly with Haruto Kimura (Stella). - [Mbed TLS FFDH improper input validation](https://xsourcesec.com/disclosures/cve-2026-34872/): FFDH key exchange lacks contributory-behaviour / peer-key validation, enabling protocol-level abuse. Found independently by Eva Crystal (0xiviel). - [Trusted Firmware program.](https://xsourcesec.com/disclosures/trusted-firmware-program/): 18 accepted findings across AMD, Intel, VMware and ARM. - [PowHSM Merkle proof bypass](https://xsourcesec.com/disclosures/powhsm-merkle-proof-bypass/): federation key management in Bitcoin sidechain. ## Research areas - [AI Infrastructure](https://xsourcesec.com/disclosures/area/ai-infrastructure/): A published CVE in the vLLM inference server and a credited advisory in AnythingLLM. Further advisories in inference servers and LLM gateways are in coordinated disclosure with their maintainers and publish with their fixes. - [Cryptography & PQC](https://xsourcesec.com/disclosures/area/cryptography-pqc/): Post-quantum signature implementations, the Mbed TLS library and the Open Quantum Safe OpenSSL provider. All credited to Eva Crystal (0xiviel), XSource Security. - [Firmware & Accelerator](https://xsourcesec.com/disclosures/area/firmware-accelerator/): SMM vulnerabilities in Lenovo BIOS, coordinated with Lenovo PSIRT: advisory LEN-220440 published 14 Jul 2026, CVE records 16 Jul 2026. Each CVE credits: "Lenovo thanks 0xiviel/Eva Crystal (XSource Security)." ## Key pages - [Continuous AI Security](https://xsourcesec.com/#services): ongoing AI red teaming and AI penetration testing for one AI application on an agreed testing schedule, with model, prompt, tool and integration changes retested within an agreed scope - [AI Security Assessment](https://xsourcesec.com/#services): a one-off 1 to 2 week assessment of a chatbot, AI agent or LLM feature, with a written report - [How it works](https://xsourcesec.com/#process): baseline, retest on change, review before release, report - [Research](https://xsourcesec.com/research/): published CVEs, advisories, upstream fixes and public audit reports - [Eva Crystal, founder and researcher](https://xsourcesec.com/eva-crystal/): background, published CVEs, upstream fixes, certifications - [Source-level audits](https://xsourcesec.com/services/): one-off audits of AI infrastructure, firmware and cryptography - [Disclosures](https://xsourcesec.com/disclosures/): every public CVE and advisory, with references - [Request an assessment](https://xsourcesec.com/#contact)
# Vulnerability disclosure for xsourcesec.com # XSource Security, Christchurch, New Zealand Contact: mailto:eva@xsourcesec.com Policy: https://xsourcesec.com/security/ Preferred-Languages: en Canonical: https://xsourcesec.com/.well-known/security.txt Expires: 2027-11-07T15:35:33Z
| – |
| 7 Nov 2026 |
| – | 7 Nov 2026 |
| – |
| 7 Nov 2026 |