Six months ago I lived outside, competing for food with my 6 brothers. Today, I have a personal TLC & Patting team, 2 passive incomes, and a personal chef. Pics: Private driver taking me to the beach. Follow me on insta (thebestestshadow) for more financial advice.

AI use is pretty wholesome: Knostic’s learnings from 13,455 publically shared AI conversations, as explored in research blog by Shayell Aharon Salomon. www.knostic.ai/blog/safe-ai...

Knostic is the Gartner Cool Vendor in AI security governance. The recognition by leaders such as the analysts behind AI security at Gartner means a lot to us. This comes right on the heels of winning the SINET award earlier this week, and BlackHat / RSAC before.

Agents are about to need a passport, one that speaks JSON and isn’t behind a paywall. Security will still need duct tape, though, to keep them running as attackers route around these stopgaps 🧵

In “Take Note: Cyber-Risks With AI Notetakers” Joe Sullivan and I write for Dark Reading on the enterprise risk landscape of notetakers. From cyber and legal to audit, AI summarizes are useful, and a time bomb🧵 www.darkreading.com/cyber-risk/t...

Knostic won again, following BlackHat and RSAC, we are a SINET16 Innovator! SINET isn't just an award. It is the practitioner's stamp of approval. From Shadow AI and oversharing to governance and defending your AI coding assistants, send me a note and let's talk!

Stop what you’re doing and read this… don’t get surprised by what’s coming. It’s time to rethink everything you’re doing on cyber defense. H/T to Gadi and Bruce for partnering on this opinion piece. www.csoonline.com/article/4069...

The upcoming AI vulnerabilities cataclysm on CSO Online, by Heather Adkins, Bruce Schneier, and myself. www.csoonline.com/article/4069...

Unpopular Opinion #1: I don’t like The Basics Not because they don’t matter. They do. I patch, I enforce MFA, I budget for backups and segmentation. But "the basics" have been turned into a cult 🧥

Every week there is yet another AI coding agent hacked, or developer compromised through it. Usually it’s about Copilot / Claude Code / Cursor. This time, it’s about Void Editor. This is today’s story, by Idan Habler, PhD: idanhabler.medium.com/d%C3%A9j%C3%...

היום ה-6 לאוקטובר הוא יום הגירית www.badgertrust.org.uk/brocktober

This wins the internet today. Via @forgebitz on Twitter

Every few years I try to make the claim that we should consider absolutely everything we see as entertainment, as any other approach is untenable. If we want to see if something is news or truth, we’d then need to specifically research it.

Like this is *insane* youtu.be/VaeI9YgE1o8?...

What's the more likely training course for developers to take, in 2027? - Programming Basics for Devs - Prompt Engineering for Devs

This only happens to you once

LinkedIn is fun

Shadow is almost 8 months old 😍 First time at the beach!

Announcing the Prompt||GTFO Bugcrowd Leaderboard episode special! In collaboration with Bugcrowd, we’re inviting top-tier bug bounty legends to step into the Prompt Pit. - 23 October, 2025 Reg: forms.gle/c7Y8Jx4YPXCf...

Someone figured out how to play DOOM over DNS... www.youtube.com/watch?v=GoPW...

Perfection

When do we see the first license that forbids MCP use? That’s what Michal Kamensky asked after I shared Code2MCP, a project that auto-wraps repos into MCP services. 🧵 arxiv.org/abs/2509.05941

When do we see the first license that forbids MCP use? That’s what Michal Kamensky asked after I shared Code2MCP, a project that auto-wraps repos into MCP services. 🧵 arxiv.org/abs/2509.05941

1/ Hi, I'm TProphet. I write the Telecom Informer for @2600.com. A lot of people have been asking me about www.nbcnews.com/politics/nat... given that I'm somewhat knowledgeable in the area. Here's my take: I'm kind of astonished that this is public, and it isn't normal that it would ever be.

This is best story by far about the SIM farm uncovered in New York by Secret Service. The intent of the farm is not, as suggested by other stories, to disrupt Secret Service protection of dignitaries attending the UN or taking out mobile service for all NY but about managing criminal enterprises

They also likely used a computer.

RTFM they say but if you read the manual and copy code examples from it you may inadvertently introduce vulns in your code 🙀 In April we audited the PHP code. Now we followed up with a review of the code snippets in PHP documentation and found 81 issues 👇 blog.quarkslab.com/security-rev...

I gave ChatGPT’s agent mode my credit card to buy groceries. Would you? Google just introduced the Agent Payments Protocol (AP2), an open standard that lets agents make payments on our behalf. 🧵 cloud.google.com/blog/product...

“Can you all please turn off your WiFi? We’re working on a trust system here.” - I ‘member. Via @nearcyan.bsky.social on the other net, through Jake Steinerman

A New season of Prompt||GTFO kicks off this Thursday (Sept 25)! Vulnerability disclosures, bug bounty leaderboard and strategy episodes, and more! - Michael Brown (Trail of Bits), @thegrugq, Daniel Cuthbert, Anton Chuvakin, Andre Gironda, and others! forms.gle/LLJMzXbh541m...

4chan is reserving flights from India to prevent H1B visa holders to get back to the U.S., following the tech industry’s instructions to employee visa holders to return to the U.S. immediately. The Internet shall remain the Internet, and Denial of Service attacks don’t have to be packet-based, and 🧵

Team Atlanta just published a 150-page report on ATLANTIS, their Cyber Reasoning System (CRS) that won DARPA’s AI Cyber Challenge, credit to Mark Griffin for surfacing this. Report: arxiv.org/pdf/2509.14589 CRS was new terminology for me. It is DARPA’s term for a self-driving security 🧵

Apparently, it’s not just artists or developers who are at risk. This paper shows how LLMs make founder selections for VC investments: arxiv.org/abs/2509.14448

I for one accept our new robotic overlords. x.com/afp/status/1...

AI-based disinformation and cyber attacks helped fuel NVIDIA’s recent drop in market capitalization, which wiped billions in shareholder value. In my years following Computer Network Influence (CNI) cyber operations, we’ve seen disinformation campaign attempts to devalue Fortune 500 stocks, 🧵

This is absolutely brilliant! AI + theology + math in a great narrative. Wow. Screenshot for a taste, via Kevin Weil. www.astralcodexten.com/p/what-is-ma...

Are you ready for this? Prompt||GTFO is back with season two, this next Thursday, the 25th. S02E01: Indiana Pit and the Raiders of the Lost Prompt. Register to watch or submit a talk, here: forms.gle/27qe2eWncrhq...

This is a level of transparency and technical lessons learned I often expect only from @Cloudflare. Interesting, and well done, @AnthropicAI. It also brings us back to my post a few days ago about the nondeterministic nature of LLMs, and how that is now changing. www.anthropic.com/engineering/...

Azure vulnerability. Take note immediately. TL;DR: Actor tokens let attackers impersonate ANY user, even GLOBAL ADMINS, ACROSS TENANTS while bypassing Conditional Access. NO MEANINGFUL LOGS exist, so attackers could fully compromise directory data with zero visibility. dirkjanm.io/obtaining-gl...

Where I can, I want to start annotating my writing on where it’s AI, and where me. As I work collaboratively with AI, the line isn’t always clear. Spitballing here for version 0.1. Here’s what I’m thinking.

A “Holy tokens, Batman!” breakthrough just dropped in RAG. Not hocus-pocus about next-generation consciousness, but pure engineering, done right, pushing boundaries and enabling enterprise adoption at scale. arxiv.org/abs/2509.010... 🧵

30 years to Hackers’ debut in Hollywood. What’s your favorite quote? #HackThePlanet

I am convinced. In the future, “Programming for Developers” will be mainstream training. It’s just like today devs don’t necessarily know asm if they code C, or JavaScript if we develop in node. They may learn Python, developing in English. But, this did make me chuckle.

A cabinet minister bound by OpenAI’s licensing, and vulnerable to prompt injection? Hobbes once imagined states as a mechanical sovereign. Albania gave that metaphor a portfolio. Reuters - www.reuters.com/technology/a... 🧵

LLMs don’t give the same answer to the same input. Until now Like a calculator that says 2+2=5, outputs drifted. Mira Murati’s team at Thinking Machines fixed it: same input = same output Another mystery of how LLMs work, solved: Remoce GPU “jitter” thinkingmachines.ai/blog/defeati...

My new robot cleaner identifies my bedroom’s closet as an extra unknown room. I’m getting Narnia vibes here. What are you hiding, “Room 7”?

I got a cold approach from an SDR selling me post-quantum homomorphic encryption... to protect AI agents.

Same week: the NHS calls an AI scribe nothing less than transformative, and a Google medical model fabricates a body part, horrifying doctors. Is AI useful, or useless? A short thread to facilitate discussion

Classic security bugs have a way of resurfacing in new domains, in this case AI. And when it is a logical vulnerability like TOCTOU, that is just cool. I get especially excited when they land in the Flowbreaking attack family (it's extra satisfying when new research validates my own work). TOCTOU 🧵