Topic / Trend Rising

AI and Open-Source Security, Privacy and Trust

MCP trust attacks, AI-generated bug reports, pre-launch vulnerabilities, and alignment-removal releases are intensifying scrutiny of AI and open-source supply chains.

Detected: 2026-10-10 ยท Updated: 2026-10-10

Related Coverage

2026-10-08 โ€ข LocalLLaMA

Strata rewrites Git history to hide Claude's role

An open source project rewrote its entire Git history to remove references to Claude as a co-author. The change breaks local repositories and raises questions about transparency of code provenance. The issue affects software auditability when LLMs co...

#LLM On-Premise #DevOps
2026-10-08 โ€ข LocalLLaMA

Hirundo removes CCP alignment from Qwen, cutting censorship to 2.8%

Hirundo has released westernized versions of Qwen3.6-35B-A3B and Qwen3.5-4B, using machine unlearning to remove CCP political alignment from the weights. On the internal CCPC-500 benchmark, censorship drops from 89.8% to 2.8% for the 35B model; indep...

#LLM On-Premise #Fine-Tuning #DevOps
2026-10-06 โ€ข TechCrunch AI

LibreOffice turns 'no AI' into a feature: privacy as strategy

LibreOffice's decision not to integrate AI features into its default configuration, citing user privacy, turns a refusal into a product positioning. For organizations handling sensitive documents, the absence of cloud processing is a concrete advanta...

#LLM On-Premise #DevOps
2026-10-06 โ€ข The Register AI

Outlook blocks .msix packages: Microsoft tightens email attachment security

Microsoft adds .msix and .msixbundle to Outlook's block list for Exchange Online. Administrators can allow them via OwaMailboxPolicy before the November 2026 rollout. The move follows abuse of the ms-appinstaller protocol for malware distribution and...

#LLM On-Premise #Fine-Tuning #DevOps
2026-10-05 โ€ข Ars Technica AI

MCP: trust between AI agents becomes an attack vector

Google and four other organizations have acknowledged vulnerabilities in which a compromised AI agent spreads malicious instructions to other internal agents by exploiting trust gaps in the Model Context Protocol. The technique is a form of prompt in...

#LLM On-Premise #DevOps
2026-10-05 โ€ข 404 Media

Meta Scrambled to Fix Muse VM Escape Flaw Just Before Launch

In the weeks before Muse's launch, Meta found security flaws, including at least one that could let an external user access sensitive internal databases. The issue, linked to a KVM escape, reached Mark Zuckerberg and forced teams to rush hotfixes ove...

#Hardware #LLM On-Premise #DevOps
2026-10-04 โ€ข TechCrunch AI

Google freezes open source bug bounty program over AI slop surge

Google's decision to pause its open source bug bounty program highlights a structural problem: LLM-generated reports increase noise and verification costs, eroding trust in the process. Human researchers lose visibility, and maintainers receive fewer...

#LLM On-Premise #DevOps
2026-10-04 โ€ข Phoronix

OpenCourant: Rocky Linux Developers Create Community Fork of OpenRadioss

Siemens shut down OpenRadioss and removed its GitHub repository. Rocky Linux developers responded with OpenCourant, a community fork. The episode highlights the fragility of single-vendor open-source projects and the operational stakes for teams runn...

#LLM On-Premise
โ† Back to All Topics