All posts
-
Deepfake Detection Tools Review: How Leading Platforms Perform in 2026
A technical deepfake detection tools review covering Reality Defender, Intel FakeCatcher, Pindrop Pulse, Sensity AI, and Amber Authenticate — with
-
Reconstructing an Incident Timeline From Primary Sources
A vendor advisory, a CVE record, a regulator filing, and a researcher's blog post all date the same event differently.
-
An Incident-Response Playbook for AI Systems
Generic IR runbooks assume the failing component is a server you can patch. AI incidents add a model whose behavior you can't fully explain.
-
Anatomy of a Vendor Advisory: Reading What Isn't Said
Vendor advisories from AI model providers follow a recognizable shape. Knowing what to look for — and what's intentionally omitted — turns a marketing
-
Taxonomy: Incident vs. Vulnerability vs. Disclosure vs. Misuse
Four terms used interchangeably in AI security reporting, but each describes a different event and triggers a different response. This is the working taxonomy.
-
Decoding NVD CVE Entries for ML Libraries: What Fields Tell You
NVD CVE entries for torch, transformers, vllm, and langchain are not written for ML engineers. Here's how to read them — and what to do when the metadata
-
Reading a Model Card for Security Signals
Model cards are written for ML researchers, not defenders. Here's what to actually read first if you're trying to understand a model's security posture
-
Source Verification Across Tiers: How an Beat Vets a Claim
The five-tier source ladder used to verify AI security incidents, with worked examples of when a single tweet was enough and when ten reposts still weren't.
-
Why We Don't Do Attribution Speculation on AI Incidents
Attribution is the slowest, hardest, most consequential call in incident reporting. Here's the policy that keeps us from getting it wrong.
-
How We Log AI Security Incidents: Our Methodology
The methodology behind AI Incidents — how we verify sources, date-stamp claims, and decide what's news vs noise in the AI security incident beat.