Oktane Preview with Harish Peri, Invisible Prompt Attacks, and the weekly news! – Harish Peri – ESW #421

Adrian Sanabria

1. NEW TECH: Drone As First Responder Programs Are Swarming Across the United States

   I remember reading some stuff on this a year or two ago and seeing folks freaking out on social media when they spot these things in the wild, but it seems truly commonplace now. With very little effort, I counted 34 police or sheriff departments that currently use commercial drone technology (most come from the EFF article here, which is over a year old now).

   Anyone that has done penetration testing, bug bounty work, or played any of the games in the WatchDog series is immediately wondering how hackable these things are. The manufacturers seem pretty limited so far:

   1. Brinc Drones
   2. Skydio
   3. Axon (acquired Dedrone)
   4. DJI

   The idea of a DFR (drone first responder) is that dispatch radio triggers a situation where they send out the drone first, before anyone else gets there (most of the time), to assess the situation. Better safety for first responders, and the ability to send the right resources to address the situation. Makes sense, but as a cybersecurity guy, I just worry about these drones making mistakes, getting hacked, getting shot at, falling on people/cars/property.

   Chula Vista, one of the first to use DFRs, claims over 1000 cases in one year where the drones led them to not send out police. Just the addition of police presence can sometimes turn non-incidents into smoking craters, so this seems like it could be a good thing. Of course, the opportunity for privacy invasion, abuse of authority, and general creepiness is high here.

2. VULN MGMT: From Rules to Reasoning: The Shift That Made Maze Possible – Maze
3. BREACHES: Eugene Man Charged in Connection with Global Hacking Network

   Dude from Alaska built his own Mirai clone out of routers, DVRs, and cameras. Like Mirai did a decade ago, he set a record for the largest DDoS attack ever (~6Tbps).

   He got busted and so did his botnet. I wonder what they do with all the compromised devices after closing a case like this?

4. DUMPSTER FIRES: Phishing training is pretty pointless, researchers find

   The research concluding that phishing training isn't worth the pain, effort, and lost trust with employees. It can hurt productivity and even hurt security, researchers have found.

   Let's not throw out the baby with the bath water just yet, though - don't conflate phishing testing with security awareness. They're often bundled in the same package from the same vendors, but these studies specifically call out phishing testing, not security awareness.

   This is the same paper we covered back in December 2024 on episode 387 (that I stupidly paid $21 for). Looks like the authors are still giving talks on it, as it's Black Hat that brought this to the forefront again.

   • Phishing in Organizations: Findings from a Large-Scale and Long-Term Study
   • Understanding the Efficacy of Phishing Training in Practice
5. NO RUSH: Microsoft Lays Out its Quantum-Safe Plans

   Microsoft aiming for 2033.

   Apparently the USGov goal is 2035.

   Might as well start now?

6. VULNERABILITIES: Breaking Into Your Network? Zer0 Effort. – DEF CON 33 Overview

   This feels a bit like Tavis over at Project Zero Day, setting his sights on a class of software and ripping it apart. Commercial ZTNA products having a bit of a rough time here.

7. SQUIRREL: Grok Exposes Underlying Prompts for Its AI Personas: ‘EVEN PUTTING THINGS IN YOUR ASS’

   I won't read the entire title, but let's say the spirit of Elon is captured well in Grok's system prompting.

Ayman Elsawah

1. APPSEC: Buttercup: Open-source AI-driven system detects and patches vulnerabilities – Help Net Security

   Adrian: I snagged a private invite to check out Buttercup a year ago, and was pretty impressed. Back then, they had some interesting lessons learned that now seem prescient. The only team during the DARPA challenge that tried to do everything with AI failed completely. It seems the way to go is to use LLMs only when you absolutely need to and use more traditional automation techniques for everything else. Very exciting that this is open source though!

Jackie McGuire

1. Zenni ID Guard – anti surveillance eyewear
2. TECH DEBT: Microsoft: August security updates break Windows recovery, reset

   Win10 is getting put out to pasture real soon, y'all!

   This will be your only reminder (maybe) from ESW

Sean Metcalf

1. AI NEWS: GenAI FOMO has spurred businesses to light nearly $40 billion on fire

   "MIT NANDA study finds only 5 percent of organizations using AI tools in production at scale"

   Adrian: I honestly think it's going to turn out to be a LOT more than $40 billion.

2. DUMPSTER FIRES: Mozilla warns Germany could soon declare ad blockers illegal
3. FUNDING: Databricks is raising a Series K Investment at >$100 billion valuation

   Series K is a lot of funding rounds

4. NEW FEATURES: Agentic Windows

   MCP, built into Windows!

   They promise it is going to be secure though, so don't worry ;)