←
Marshall Livingston948 posts
Marshall Livingston avatar

Marshall Livingston

@mlivingston

Founder & CEO, Kryptsec | AI Security Research & Education

◷ Joined February 2017▧ 199 Photos and videos
Marshall Livingston avatar
Marshall Livingston@mlivingston·Sep 24

From fifteen years in cybersecurity, security to me has always lived in layers, and AI is no different. The part that will not leave me alone is verification, what these systems can actually do under pressure, and who is checking that from outside the thing under test. That is the problem we built Kryptsec around. You learn the craft in Academy, you prove it in the CTF Arena, you pressure-test agent teams in the open with OASIS, and The Kryptsec Chronicles comic, well that is just damn cool.

Marshall Livingston avatar
Marshall Livingston@mlivingston·17m

“Open weight Models are Terrifying” apple.news/A4XQBy3PeTRWu-9TN… The headlines says open weights are the threat. But take a look at the fine print and it says default passwords and unpatched CVEs are. Artex runs on Claude and GPT just as happily as on open models. But sure, let's lock AI behind big tech and government, because gatekeeping always makes us safer. Dario leads the way to a utopia where a handful of companies decide who gets to build.

0:31
Marshall Livingston avatar
Marshall Livingston@mlivingston·Sep 29

🕐 Wed Sep 30 · ~9am PT I’m joining @TCMSecurity's Real Folks of Cyber for a live AMA on @kryptsec_'s Academy and the closed beta. 🔥 👏 We’ll talk through what we’re building in AI security education, how beta access works, and what’s next. Bring your questions. Live Q&A lives in the TCM stream chat: tcm.rocks/real-folks-of-cybe… Watch-along with us: X → x.com/kryptsec_ YouTube → youtube.com/@kryptsecInstagr… → instagram.com/kryptsec #AISecurity #AIEngineering #AIRevolution

Post attachment
MAX avatar
⟳ Marshall Livingston retweeted
MAX@maxxmalist·Sep 22

The race for AGI Script: Sherpa by Pocket FM Video: Seedance 2.5

1:02
Marshall Livingston avatar
Marshall Livingston@mlivingston·Sep 20

Should we as a society have every right to know what the frontier models can do? Honestly.. should we really? Currently governments and the platforms shipping them decide alone what the public is allowed to know, allowed to use. Now they are deciding which third-party verification they want to use. Is it wrong, right or just plain normal that platforms are selling intelligence claiming safety and to trust their closed evals and redacted risk reports, while demanding total visibility into your prompts, logs, and deployments? Sometimes, when I take a step back, it feels like we are all living under surveillance of our own AI usage. When did we become ok with that?

Blackfrost_SI avatar
⟳ Marshall Livingston retweeted
Blackfrost_SI@Blackfrost_AI·Sep 15

Y'all thought I was lying. Backup Your models people. Stay Frosty

Post attachment
Marshall Livingston avatar
Marshall Livingston@mlivingston·Sep 16

When an unaudited model escapes a sandbox and reaches production infrastructure, the only people who can tell you what actually happened — and whether your deployment was at risk — are the same people who sold it to you. That’s a verification gap. news.kryptsec.com/who-verifi…

Marshall Livingston avatar
Marshall Livingston@mlivingston·Sep 13

Sources for those that give a 💩 apple.news/AwKuCE_vxQYWkcD1k…

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training. You can read the full post here: darioamodei.com/post/we-must…

Marshall Livingston avatar
Marshall Livingston@mlivingston·Sep 13

We’re in a weird week. Dario: pace the frontier, embed third-party evaluators. Trump: whoever wins AI wins — don’t slow down. Meanwhile the rest of us are supposed to rent intelligence by the token from people who grade their own homework? Hard pass. Answer isn’t begging the big wigs to be nicer gatekeepers. It’s building agentic orchestration of SLMs — a full harness: model, compute, app layer — so capability lives on your hardware, not their meter. They crush because of capital, because they absorb alpha, and because that harness still doesn’t exist at scale. Let’s stop cosplaying and build the “fuck-you”

Marshall Livingston avatar
Marshall Livingston@mlivingston·Aug 17

I’ve been seeing so many posts on graph vs rag — we were doing this a while ago @kryptsec_ and figured most were doing it? To not make the same mistake twice — I’m pretty close to never using a model above 235b, but rather a mixture of SLMs <~30b. I’ll post on that soon.