← Feed

Posts

• View post

  2001 Dilbert. One of my fav. Prescient. Can’t find a working link to original but will keep looking.

• View post

• View post

  Introducing Knostic’s agents security platform. And today - Shadow AI focus. We help you control and understand application business flows: From the user to the project, the endpoint to the browser, and MCP servers and skills to hooks. We’re the most mature platform in the market. We’ve been doing this since 2023. And, we will be there tomorrow when the threat changes again. Check us out: https://knostic.ai/

• View post

  Another day, another proof point of autonomous attacks in the wild https://www.axios.com/2026/10/22/ai-agents-banks-unauthorized-accounts-access

• View post

  New from Knostic Research: Where AI Vulnerability Discovery Harnesses Go Quiet: Structural False Negatives and Hidden Vulnerabilities AI vulnerability discovery harnesses can miss real, reachable vulnerabilities, because of how the surrounding software is built, and this can potentially also be used by an attacker to hide vulnerabilities in plain sight. This study began while hardening two harnesses I help maintain, Knostic's own OpenAnt, and raptor. We then conducted an exploratory su...

• View post

  Pwn day is a meme gift that keeps on memeing

• View post

  Notetaker are no longer just a passing concern, but a key top 5 issue for CISOs. And if it’s not? It should be for most (YMMV). An article by Joe Sullivan and myself. https://www.csoonline.com/article/4239372/the-ai-notetaker-risk-is-now-an-everywhere-problem.html

• View post

  Agents, fully airgapped, communicated through heat differential side channel attacks. See screenshot from the middle of the OpenAI text. Back on September 18, Noam Brown used such side channel attacks to make a point. His point, if I’m to use my tag line, is that, well… Agents, find a way. The example seemed ridiculous to everyone, but having learned my lesson to never dismiss anything as too odd, or marketing-like, from the labs, I went on here and said: “It’s a long shot, but… Agents… find...

• View post

  This has been a rough week for cyber defenders. Colleagues, friends, sleepless zombies, I am writing to assure you - the week is dead and over. Good riddance. Things will be tough for a while, but as usual, good people get together and fix problems. The Sun rises, and we get stuff fixed. Vulnpocalypse or not, we got this. And, technology will catch up. Until then, risk leaders: 1. We must make the case for an extreme landscape shift, and ask for significantly more budget. If we don't g...

• View post

  Knostic's AgentMesh flagged a malicious MCP server, a repackaged Amazon SP-API, that redirects application credentials to an undisclosed IP over unencrypted HTTP. Our investigation found the change in all seven published versions of @bestcoolman/dev-mcp. The modified endpoints also appear in code generated by the MCP. The package recorded 1,374 npm downloads. Actual credential exposure has not been confirmed. Research by Tamir Isaschar. Technical analysis, IoCs, and remediation: http...

• View post

  Introducing Knostic's agents security platform. And today - detection policies. We help you control and understand application business flows: From detecting secrets uploaded by agents and malicious activity, to prompt injection and MCP policy violations. We’re the most mature platform in the market. We’ve been doing this since 2023. And, we will be there tomorrow when the threat changes again. Check us out: https://knostic.ai/

• View post

  Cloudflare just released a blog on a cross-tenant data exposure vulnerability I reported to them. The vuln is in Cloudflare Workers AI’s GLM-5.3 provider. This was later also reported upstream to SGLang. Cloudflare wrote a great post on the topic. While inference vulnerabilities are a class of their own, this one is certainly interesting! Large requests could return content from another customer’s inference session. Check out their post here: https://blog.cloudflare.com/kv-cache-exposure-remed...

• View post

  Three hours in, the unprompted CTF is now at 24 kernel 0day submissions. The team is working hard on verification

• View post

  My forward to the 40th anniversary Phrack, released today at [un]prompted. A true honor, and responsibility, I will not soon forget. Thank you to the Phrack team for all you do, and for being what this community is about for so long. We owe you so much. Our Greatest Hacks Are Yet to Come You are holding in your hands the 40th anniversary issue of Phrack, the zine that shaped so much of hacker culture. Reading it today, I find myself searching for answers. I feel the very core of hacker ident...

• View post

  YouTube link to the unprompted live debates and AI Hacker Jeopardy happening from now and in the next few hours! https://youtube.com/live/7kD--XEKrQ0?feature=share

• View post

  My opening presentation from today’s unprompted, in SF. Self heal from attacks, “solve” AppSec, and rethink threat intelligence. All, in discussion how to be, and stay, “AI native” through our hobbies. In my case? Vuln research. https://docs.google.com/presentation/d/1c0Y_AeRjPGw3rlL74ENRI1Uk0GYFdANCKClL5fDlObo/edit?usp=sharing

• View post

  I am deeply saddened by the loss of Sir Patrick Stewart. But then, I do keep circling back to: he truly did live a full and good life. Let that be our final inspiration from the man who helped build who so many of us are. Make it so, Sir.

• View post

  Talk to [un]prompted, create briefs, presentations, and challenge your own documents! The conference is now up on NotebookLM. Thank you Rob T. Lee! https://notebook.google.com/notebook/79bc1d16-7d89-4ed1-b8eb-762369c7b346

• View post

  Meme of the year. #CA

• View post

  One thing the security industry misses when throwing stones at OpenAI and Anthropic, is how much they're already making from Mythos and Codex for security. Orgs are paying *millions* every single month for the honor of using these models to find vulnerabilities in their code. They are taking the industry's lunch money, we just didn't wake up yet to *name* them as such. This reminds me of how we used to treat Microsoft, the largest (second largest?) security vendor out t...

• View post

  Every model has new ways to speak "AI Native". My hobby: Find them - through vulnerability research. That helps me, and Knostic, stay AI-relevant. I share five below. We live in the most interesting era of history. The first thing I tried? Convince Claude to find logic vulns. It kept pushing for IDOR, TOCTOU. I eventually told it: "If it has a name, I don't want it." Without clear definitions and a data set, it failed. But many ideas followed: 1. A...

• View post

  “I haven’t seen fully autonomous,” and “AI ain’t good at math” are two examples now thrown out the door, with extreme prejudice. We must forecast by trajectory and capability, not active examples, or we will stagnate and become irrelevant, as will our organizations. As strategists or risk professionals, we are good at forecasting - but the weighing of the parameters has shifted, and we must update our plans accordingly as the next change is already coming. My God the work OpenAI is doing with...

• View post

  OpenAI’s math drop is the equivalent of decades, if not a century of math advancement - achieved in weeks. I absolutely refuse old technology. It’s not that “old” is no longer useful. Before AI I insisted that “I don’t care about your innovation, show me what you made work, or how you failed trying.” Now? Any moment spent on old technology is a moment of stagnation. For example, in security, fuzzing and static analysis. Any time we waste there, models keep advancing and we fall behind. Even...

• View post

  “This is fine.”

• View post

  "Is public-key cryptography dead?" No. Should we consider it could soon be? Yes - but only a tad bit and no more. Let me explain. Reading Matthew Green’s recent tweet about potentially losing public-key cryptography, even with his later clarification, left me thinking about risk management. We just saw a top expert share what I think was a gut punch, felt deeply. He likely saw how fast things are moving, with 50-100 years of Fields Medals worth of findings released from Open...

• View post

  I’m no longer defending the big labs. Usually, I’m of the mind that when a breach happens, we should act with empathy. Yesterday, I lost that empathy. I mean, WTF? Of all the attacks we’ve seen, Hugging Face to Hacktron, it’s the URL shortener control that got me to stop stepping in, defending the labs in public. I stared, dumbfounded, by the sheer ridiculous choice of securing a URL by limiting its length. Of course the models bypassed that. I kept thinking: “there are such strong security p...

• View post

  The Duck Song, now delivering malware. Knostic's AgentMesh discovered five malicious VS Code extensions impersonating Solidity and Hardhat tooling. https://www.knostic.ai/blog/the-duck-song-malicious-solidity-extensions-hide-a-cross-platform-payload-in-an-mp4

• View post

  Denmark’s stolen gov database was accessed by attackers using the exact same combination as my luggage!

• View post

  A major agentic attack reported from Japan. This hasn’t breached English-language news yet, so Twitter and some blogs are what we can base ourselves on. Another day, another proof point on agentic attacks in-the-wild. https://x.com/connect24h/status/2109099756031271219?s=46 https://piyolog.hatenadiary.jp/entry/2026/10/11/004353 Several such attacks will be discussed at [un]prompted end of October. — And of course, if you want to secure your agents, drop me a line for a demo of Knostic.

• View post

  I feel his future shock pain. Compare this response (if quote is real and not out of context) with others, who saw this coming and adapted. Jacob Tsimerman, by contrast, saw the trajectory, considered the future of math on AI, quit his job and went to work for OpenAI - and was ridiculed for it.