David Kaplan

@depletionmode

Security Research. Opinions and private research are my own Lover of all things JSR $F7D7 💪🇮🇱 עם ישראל חי

Normally Closed

depletionmode.com

Replies Reposts Media

David Kaplan
@depletionmode
Sep 23
Software was (is) so awesome!
Jimi Plouffe
@JimiPlouffe
Sep 22
The masculine urge to fire up Visual Basic 4, drag a few buttons onto a form, hook it up to a 33.6k modem, and casually build ChatGPT for Windows 95.
1
David Kaplan
@depletionmode
Sep 20
Awesome stuff! Love the cool unlock animation as well. I had to get my agent to make some tweaks to get working on my system (PAM orders to allow for 0-sized passwords on the lock screen, disabling the glare cue because it was no good on my webcam and one or two other things.
Ayan 🔳
@ayandexyz
Sep 19
Created Face unlock on Omarchy, with liveness detection 👀 - Look at your laptop → unlocked. No password typed, no keystrokes injected - Defeats printed photos & phone screens (5-cue liveness model) - Goes through PAM, same path as sudo/hyprlock. Zero passwords stored -
00:00
David Kaplan
@depletionmode
Sep 17
Agents as confused deputies is a recurring pattern - and there are some weird/adjunct forms of this behaviour that is emergent in our own trace data. What we found in the case described in Sam Lego’s blog post below is that Claude’s MCP connection becomes a type of ambient
Origin
@originhq
Sep 17
Replying to @originhq
Here’s one example we found while reviewing agent activity. Codex launched Claude Code to use a tool it didn’t have, then disabled Claude’s approval checks. Read the investigation. originhq.com/blog/codex-lau…
3
David Kaplan
@depletionmode
Sep 17
@typesafeai 's Jev is awesome. I've used it to create a custom/dynamic MCP server framework that can turn any site (including ones requiring auth) into an extremely fast MCP. * mcps isolated in docker instances * Jev drives headless chrome * agent skill - just instruct as to
00:00
3
David Kaplan
@depletionmode
Sep 16
Ok, got access to TypeSafe - tried it out for 5 minutes on a simple use-case we have in-product at @originhq. First impressions: I'm seriously impressed.
Diogo Almeida
@CompleteSkeptic
Sep 15
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x
00:00