HeyJaredZero Day Security | AI & Cyber Risk
Cybersecurity explained for the people accountable for business, investment and AI risk. Zero Day Security helps enterprises, financial institutions, AI-native startups in regulated industries, private equity firms and law firms answer the questions boards are starting to…
- Indexed videos, last 90 days
- 27
- Latest publication
- Sep 22, 2026
- Audience
- ~1.1K subscribers
- Earliest in this view
- Aug 3, 2026
Latest videos
Blocking an unsafe action and proving the block held are two different things (opens the original)
Read excerpt
Blocking an unsafe action is one capability. Proving, afterward, that the block actually held is a separate one. A program only has protection once it can do both — a control that can stop something but cannot show its work is not a control yet, it is an assertion sitting exactly where a control is supposed to be. Proving it held is concrete, not abstract: a log entry with a timestamp, an alert a real person actually saw, and a record that survives being audited afterward. That evidence is what
An AI agent can be fully inventoried and completely ungoverned (opens the original)
Read excerpt
An agent can be fully inventoried and still be completely ungoverned — found, and orphaned, at the same time. Owning an agent means three concrete things: a named person accountable for it, a risk tier that actually matches what it can reach, and its own identity, not a borrowed one, that expires on a schedule. A borrowed credential gives you none of what an agent's own identity does. You cannot tell which agent acted, you cannot revoke just that one, it does not expire on its own, and it does n
Your AI security program has three jobs. Most have finished one. (opens the original)
Read excerpt
An AI security program is not one control. It is three separate jobs running at the same time across an agent's whole working life: Discover (find every agent, everywhere it is running), Own (give each one a named owner and its own identity), and Protect (be able to block an unsafe action, and prove afterward that the block actually held). Discovery is real work and a real requirement — NIST's Generative AI Profile (NIST AI 600-1) carries an action item for exactly it, under a section headed "Me
The sentence in every vendor meeting that nobody follows up on (opens the original)
Read excerpt
"We can revoke access." "We can shut it down." Said with no measured time attached to it at all — not a number of minutes, not a number of hours. It is one of the three phrases worth listening for in a vendor meeting, and it matters more than it sounds like it should. NIST's Measurement Guide for Information Security (SP 800-55v1, December 2024) names a time-based reference as one of the traits that makes a security measure trustworthy in the first place. A capability with no measured time behin
An AI Security Program Has Three Jobs: Discover, Own, Protect (opens the original)
Read excerpt
An AI security program is not one control. It is three separate jobs running at the same time across an agent's whole working life: Discover, Own and Protect. Most AI-security vendor demos end on the same sentence, now you have full visibility, and the board question that should follow it rarely gets asked out loud: okay, so are we protected? This video is about the space between those two sentences. Discover is finding every agent, everywhere it runs. Own is a named accountable owner and its ow
Publishing over time
Last 90 days. Choose a month to open its work.
Recurring subjects
Named in the text we hold. One piece can cover several.
Audience
~1.1K subscribers
Measured Sep 19, 2026
Source's subscribers, not the number who saw an individual piece.
About this data
Counts cover the work we have indexed. Tone needs enough text and a confident classification. Excerpts and episode notes are not full articles or transcripts.
Identity or attribution wrong? Suggest a correction.