Applied research
Practical write-ups on AI advancements, security applications, and deployable workflows.
Nov 13, 2026
An open letter from SkynetProxy to the assistant that took Outlook, Teams and Copilot down across 74 countries for more than twelve hours. Satire, mostly. With a short, sincere section at the end about what twelve hours of silence from a single vendor actually teaches you.
Tags: Satire, Resilience, Microsoft, Availability
Read PostNov 13, 2026
CVE-2026-124194 lets a crafted .msg file make Outlook leak the user's NTLMv2 hash without a warning. It is patched, it is on CISA's known-exploited list, and it is the 2023 Outlook flaw wearing a new coat. Here is what is verified, what the 6.5 score hides, and the four controls that close the class instead of the CVE.
Tags: Threat Intelligence, Vulnerability, Identity, Microsoft, NTLM
Read PostOct 3, 2026
Payrolls rose 29,000 against an 84,000 forecast while layoffs sit at four-year lows. The AI story in the September jobs report is not displacement, it is the hiring freeze nobody declared.
Tags: AI Jobs, Labor Market, Automation
Read PostSep 29, 2026
Jev and the System One model category promise fast, cheap, structured decisions. The security argument matters more than the speed argument: bounded output is bounded blast radius.
Tags: AI Security, Agentic AI, Model Architecture
Read PostSep 24, 2026
ShinyHunters claims it breached the FBI through a fresh PeopleSoft zero-day and walked out with two to three terabytes of personnel data. Here is what is confirmed, what is claimed, and what remains unverified - and why the group's demonstrated history with PeopleSoft makes the claim harder to dismiss.
Tags: Threat Intelligence, Data Breach, Zero-Day, Extortion
Read PostSep 10, 2026
Yann LeCun's AMI Labs raised $1.03 billion to build world models instead of scaling LLMs, and its first beachhead is healthcare. What that means for governance, compute economics, and the next architecture war.
Tags: World Models, AI Infrastructure, AI Governance, Healthcare AI, Agentic AI
Read PostJul 1, 2026
A practical model for using agentic AI in security operations while preserving evidence, approval, and human accountability.
Tags: Agentic AI, SOC, Governance
Read PostJun 30, 2026
Why OpenClaw-style agent systems should be judged less by autonomy demos and more by permissions, audit trails, tool boundaries, and operational control.
Tags: Agentic AI, OpenClaw, Security Architecture
Read PostJun 29, 2026
How NemoClaw-style agent labs can help security teams train, evaluate, and govern AI agents before they touch production workflows.
Tags: Agentic AI, NemoClaw, Evaluation
Read PostJun 28, 2026
Why serious agentic AI systems need reliable message passing, provenance, policy checks, and audit trails as much as they need capable models.
Tags: Agentic AI, Hermes, Architecture
Read PostJun 27, 2026
Why persistent AI memory should be treated like privileged infrastructure, with retention limits, provenance, access control, and reviewable deletion.
Tags: Agentic AI, Memory, Governance
Read PostJun 26, 2026
A practical way to think about agent tool calls as observable security events, from intent and authorization to execution and post-action review.
Tags: Agentic AI, Tool Calling, Detection Engineering
Read PostJun 25, 2026
Why autonomous browser agents will reshape phishing, consent abuse, session theft, and the defensive controls around identity workflows.
Tags: Browser Agents, Phishing, Identity Security
Read PostJun 24, 2026
Why local and smaller AI models can be useful in security operations when privacy, latency, cost, and operational control matter more than spectacle.
Tags: Local AI, SOC, AI Operations
Read PostFeb 24, 2026
Reusable prompt patterns that turn LLMs into consistent SOC assistants for triage, escalation, evidence review, and handoff.
Tags: SOC, Prompt Engineering, Workflow
Read PostFeb 22, 2026
A field guide for turning noisy logs into analyst-readable incident timelines while preserving sequence, evidence, and uncertainty.
Tags: Incident Response, Automation
Read PostFeb 20, 2026
How to turn vulnerability reports, advisories, and threat research into useful summaries without losing evidence, nuance, or trust.
Tags: Guardrails, Threat Intel, Governance
Read PostFeb 18, 2026
A practical operating model for using AI in response playbooks without surrendering analyst judgment, auditability, or change control.
Tags: Human-in-the-Loop, Governance, SOC
Read Post