PREVIEW. This is how it will look in an inbox; it was not sent to you. Each real email carries its own unsubscribe link and Keryn's postal address. KERYN HEADLINES Daily, Tuesday 6 October 2026 Retrospective Covers 5 October 2026 00:00 to 6 October 2026 00:00 UTC Web: https://keryn.dev/headlines/daily/2026-10-06/ TOP DEVELOPMENTS [1] MALFEX npm Malware Campaign: Three Payloads And An Adversary That Signs Their Work Checkmarx · Incident · https://checkmarx.com/zero-post/malfex-npm-malware-campaign-three-payloads-and-an-adversary-that-signs-their-work [2] VA-26-278-01: Canimaan Software ClamXAV local privilege escalation CISA · Advisory · https://github.com/cisagov/CSAF/blob/develop/csaf_files/IT/white/2026/va-26-278-01.json [3] Safeguarding LLMs via Model-Agnostic Latent Safety Signals from Dark Knowledge arXiv · Research · https://arxiv.org/abs/2610.07532 SCA [4] Backdooring Sparse Autoencoders arXiv · Research · https://arxiv.org/abs/2610.06049 AGENTIC SDLC [5] Correct Code, Broken Contributions? SWE-CC: Benchmarking Repository Policy Compliance for Coding Agents arXiv · Research · https://arxiv.org/abs/2610.06193 [6] Where Did the Repair First Go Wrong? Localizing the Origins of Silent Failures in Agentic Vulnerability Repair arXiv · Research · https://arxiv.org/abs/2610.06163 [7] ReviewBench: An open benchmark for AI code review GitHub · Analysis · https://github.blog/ai-and-ml/github-copilot/reviewbench-an-open-benchmark-for-ai-code-review AI-BOM [8] How Much Assurance Does Your AI Supply Chain Need? CoSAI · Analysis · https://coalitionforsecureai.org/how-much-assurance-does-your-ai-supply-chain-need REMEDIATION [9] New Dell System Update flaw lets hackers gain root privileges BleepingComputer · Analysis · https://bleepingcomputer.com/news/security/new-dell-system-update-flaw-lets-hackers-gain-root-privileges [10] Microsoft Exchange Flaw Lets Authenticated Attackers Read Other Users' Mailboxes The Hacker News · Analysis · https://thehackernews.com/2026/10/microsoft-exchange-flaw-lets.html [11] Realtek Jungle SDK Exploit Attempts Deliver Cling Botnet With STUN-Based C2 The Hacker News · Analysis · https://thehackernews.com/2026/10/realtek-jungle-sdk-exploit-attempts.html TRUST [12] Can Power Draw Constrain Covert Compute? Limits of Analogue Verification for AI Governance arXiv · Research · https://arxiv.org/abs/2610.07476 [13] Efficient Auditing of Adversarial AI Agent Behavior from Agent Traces arXiv · Research · https://arxiv.org/abs/2610.07256 [14] ANT: A Multi-Granularity Network Traffic Dataset and Benchmark for Agents Behavior Auditing arXiv · Research · https://arxiv.org/abs/2610.06514 MODELS [15] Understanding and Enhancing Backdoor Persistency in LLM Agent Post-Training arXiv · Research · https://arxiv.org/abs/2610.07510 [16] Dynamic Budget Allocation for LLM Evaluation under Hard Resource Constraints arXiv · Research · https://arxiv.org/abs/2610.07362 [17] Jailbreaking Open-Weight LLMs via Random Embedding Perturbations arXiv · Research · https://arxiv.org/abs/2610.07125 AI SECURITY [18] MCP for agent-to-agent comms may be the riskiest protocol you've never heard of Ars Technica: Security · Analysis · https://arstechnica.com/security/2026/10/vulnerability-in-agents-from-google-and-others-exposes-structural-flaw-in-mcp [19] Lineage-Aware Memory Governance: A Derivation-Gated Framework for Privacy-Preserving Column-Level Access Control in Enterprise AI Agents arXiv · Research · https://arxiv.org/abs/2610.07258 [20] South Korea probes bank breaches amid suspected AI-powered attacks BleepingComputer · Analysis · https://bleepingcomputer.com/news/security/south-korea-probes-bank-breaches-amid-suspected-ai-powered-attacks ADVISORIES IN THIS WINDOW 161 npm advisories · 132 critical · 14 high · 13 moderate · 2 low 42 RubyGems advisories · 42 critical 17 PyPI advisories · 3 critical · 7 high · 7 moderate 4 crates.io advisories · 2 high · 2 moderate 3 Multi-ecosystem advisories · 1 critical · 2 moderate 2 Packagist advisories · 2 moderate 2 Go advisories · 2 moderate All advisories: https://keryn.dev/headlines/advisories/ Titles, links and short excerpts only; every item links to its source. Preview: the unsubscribe link and Keryn's postal address are added to each email when it is sent.