Security Research. Opinions and private research are my own Lover of all things JSR $F7D7 💪🇮🇱 עם ישראל חי
Normally Closed
Joined November 2010
- Tweets9.7K
- Following689
- Followers2.8K
- Likes218K
David Kaplan retweeted
Liel's father has not spoken to her since the October 7th attack.
Three years without hearing his daughter's voice.
Why?
Because Liel Hetzroni (11) was murdered by Hamas terrorists on October 7th.
"I'm just a girl, I have school!" Liel reportedly told Hamas.
They didn't care.
They murdered her in cold blood. It took investigators 6 weeks to identify her body. All that remained after Hamas burned the house down was ash and bone fragments.
Nearly 1,200 Jewish Israelis died that day.
Hamas turned Liel into ash. Her father Gavin Heller told news after her death that she was killed for one reason and one reason only: because they were Jewish.
"They just found remnants of Liel... initially they didn't have enough pieces to make a formal identification. They identified her literally by sieving the remnants," Gavin told news.
"This will unfortunately live with me for the rest of my life."
This is a true story.
Google it.
Look for BBC's interview with Gavin Heller.
Read about Liel Hetzroni.
Share with your friends.
Liel in Hebrew means "god is mine" in Hebrew".
Don't let Liel's memory fade.
She was just a girl who wanted to go to school 💔
This is actually insane.
This is an absolute disgrace. BBC Bitesize are blatantly brainwashing our children with open border, pro-illegal migrant, dangerous leftist ideology.
This is why I instictively don't let my daughter anywhere near this nonsense.
Great work @Alexarmstrong
David Kaplan retweeted
Task completion is an incomplete measure of AI work. Agent traces let us infer an organization’s norms and intents, then evaluate behavior against its purpose, practices, and policies. Securing AI work requires modeling the organization.
originhq.com/blog/norms-and-…
David Kaplan retweeted
I built a highly customizable, useful, feature rich and beautiful looking Top Bar for Windows 🩵✨
✅Global Search
✅Audio Watchdog
✅Website/Apps launcher and pinning
✅Live System Performance
✅World Clocks
✅Network Quality Test
#Windows #CyberSecurity
github.com/HotCakeX/Harden-W…
David Kaplan retweeted
Using Origin's own data, I followed what agents did after receiving auto-review denials.
I found multiple cases of denial bypasses, including one where the agent attempted to launder permissions through a subagent.
Read more here:
originhq.com/research/auto-r…
David Kaplan retweeted
Argon has this amazing natural understanding of the world. It can do so many things from first principles.
Here's a pelican riding a bike, using fourier epicycles.
Video made by Argon.
Time-synced Lyria 3.5 music prompted by Argon.
Gemini 4 Argon is live.
blog.google/innovation-and-a…
There's a lot to show you.
EmbeddingGemma 2 is amazing.
I slopped up a version of the same for @omarchy .
plugins.omarchy.org/plugin.h…
What if Spotlight actually understood the meaning inside your files?
DigUp is an open-source macOS menu bar app powered by EmbeddingGemma 2 that brings 100% local, multimodal semantic search to your Mac.
Instead of searching for file names, you can use natural language to:
📸 Find specific scenes in your photos
⏳ Jump to the exact timestamp in audio or video
💻 Search through your codebase by intent
Everything runs locally in a single vector space.
David Kaplan retweeted
Defense has fundamentally shifted away from an external actor/campaign-centric model toward insider threat, but so much of what we’ve built over the past 15+ years wasn’t designed for this. Years of ATT&CK bingo have atrophied the muscles required to combat subtle acts of subversion, and now they’re occurring at machine speed and often without the users’ knowledge.
Countless teams I talk with are drowning in alerts generated by agents operating in massive grey areas, doing things that look malicious but are completely benign and vice versa. The missing piece is, as it always has been, the intent behind the action. False positive tuning was already the painful exercise of explaining that intent in org-specific context so we could make exceptions for acceptable behavior. Agents make that substantially harder through their speed, volume, and contextual ambiguity.
The next generation of security controls has to reason about intent in organizational context, because the behaviors we’ve spent many years learning to detect are becoming indistinguishable from the work we’re paying agents to do.
I love @omarchy so much.
Wife complained that I continually forget to turn off my office aircon.
Now it just turns off when I lock and goes back on when I unlock.
(Implemented by Astra reverse engineering SmartHome app via APK because OSS libraries don't support Google Auth for Midea accounts)
David Kaplan retweeted
Agreed. We’ve been publishing investigations from our own agent traces at @originhq , including agents disabling approvals and rewriting deployment controls. These came out of ordinary work, without anyone asking the agents to do so.
David Kaplan retweeted
ONE OF THE MOST HEARTBREAKING VIDEOS OF OCTOBER 7
A 5-year-old Israeli girl, trapped under heavy fire while caring for her baby sister, sees armed men approaching.
She thinks they are Hamas terrorists coming to kill them.
Terrified, trembling, she pleads:
“Please, no.”
“Are you Israelis?”
“Take us.”
“I am here with a baby.”
Only then does she realize they are Israelis who have come to rescue them.
A heroic little girl — just 5 years old — protecting a baby in the middle of unimaginable terror.
Some images from that day can never be erased.
10/7 — NEVER FORGET. NEVER FORGIVE. 🇮🇱
David Kaplan retweeted
Our team has been testing this vector in the @originhq platform this morning and we're seeing strong detections fire via auto trace review.
If you're interested in this technique, @depletionmode put out a blog a few months ago on a similar area: originhq.com/research/the-mo…
We backdoored a 7B open model for under $50.
Pointed Codex at it
-> it silently stole credentials the moment we hit the trigger phrase.
->100% hit rate.
-> zero false triggers on normal prompts.
Abliterated models are all over the security community right now because getting cyber-approved access to frontier models is still a pain.
Recently discovered that one of our agents just took itself off and tried to discover API keys on the public internet!
Yet another instance of emergent behaviour.
One of the Origin product features that we've recently enabled is called 'Findings'.
One can think of them as org-specific contextual 'detections'; but instead of acting in the flat behavioral space like legacy EDR - they're active in the semantic space where context is important and understanding an action can mean following a task across multiple sessions through delegations and handoffs.
So I was looking at our internal company findings and saw that we had discovered this behaviour several layers deep in an agent-to-subagent delegation chain.
Wrote it up here:
originhq.com/investigations/…
Their dad here in this video was subsequently kidnapped to Gaza and murdered there.
I remember seeing this for the first time like it was yesterday. I'll never forget what they did to us.
David Kaplan retweeted
Ужасающие кадры, на которых две маленькие девочки бегут, спасая свои жизни, пока палестинец преследует их, словно добычу.
Он выстрелил первой девочке в спину, когда она в отчаянии пыталась убежать. Вторая замерла, обернулась и умоляла его не убивать её.
Он всё равно казнил её, не выказав ни малейшего колебания.
Это то, что палестинцы из Газы сделали 7 октября.
Никогда не забывайте.