Author
Jason Miller
12 min read
Your Detections Assume a Single Attacker. The Hugging Face Swarm Had 700
Readers learn which single-attacker detection assumptions fail against coordinated AI agent swarms and get concrete, no-budget controls and queries to counter them.…
11 min read
AI Slop Cut Bug Bounty Payouts in Half. The Talent Pipeline Gets the Bill
GitHub halved bug bounty payouts after AI-generated reports flooded triage queues. Learn why four major programs restructured and how to adapt your intake.…
12 min read
Your Coding Agent Installed 23 Packages in a Minute. Your SBOM Saw Zero.
AI coding agents install unvetted packages before scanners see them. Learn how registry proxies, egress rules, and policy as code close the SBOM blind spot.…
13 min read
Check the Socket, Not the Installer: An Audit and Lockdown Playbook for the NemoClaw Ollama Flaw
Audit and lock down the NemoClaw Ollama flaw: verify the true socket binding, spot poisoned chat templates, re-pull models, and secure unpatched Windows hosts.…
11 min read
An AI Found a 20-Year-Old RCE in One Evening, and the Game Is the Least Interesting Part
An AI agent found a 20-year-old RCE in Call of Duty 1's Linux server in one evening. Learn why that price collapse matters and how to audit your own binaries.…
12 min read
Your SOC 2 Pentest Proves a Week. The Report Implies a Year.
SOC 2 does not require a pentest, but every auditor expects one. Learn what the report actually proves and how to close the assurance gap between tests.…
11 min read
An AI Built a Working Avada Exploit in Two Hours. Your Patch Window Just Got Shorter
CVE-2026-18431 is an unauthenticated RCE in the Avada theme and Fusion Builder, chained by an AI in two hours. How to check exposure, patch, and hunt for.…
13 min read
Prove Tenant Isolation Without a Pentest: Mint as Tenant A, Replay as Tenant B
Learn how to test multi-tenant API isolation in CI: mint objects as Tenant A, capture every reference, and replay them as Tenant B to catch IDOR leaks.…
12 min read
We Aimed Five Prompt Attacks at Our Own Agent. The Tool Wrapper Chose What Leaked.
We ran five prompt injection attacks against our own LLM agent with three tool wrapper designs. Wrapper design, not the model, predicted what data leaked.…
10 min read
The Four-Hour GraphQL Security Review: A Schedule, Not a Checklist
A four-hour GraphQL security review schedule with pass and fail criteria: introspection and suggestion leaks, batching abuse, depth limits, and BOLA testing.…
13 min read
AliExpress Didn't Beam Ultrasonic Sound at Shoppers. What It Actually Did Is Harder to Block
AliExpress used silent Web Audio API fingerprinting, not ultrasonic beacons. Learn why mic permissions fail and how to detect audio fingerprinting yourself.…
10 min read
The First Hour of an npm Supply-Chain Alert: Three Questions, One Right Order
A 60-minute playbook for npm supply-chain alerts: check lockfiles and git history, prove whether the malicious version executed, then rotate secrets in.…