
Sr. Security Engineer @Google Threat Intelligence group embedded with @GoogleDeepmind. Opinions are terrible and uniquely mine.
New York, NY
Joined September 2021
- Tweets309
- Following183
- Followers163
- Likes103
Pinned Tweet
If you're at @BlackHatEvents Europe today, come check out the launch of SaferPickle, our framework for preventing unsafe pickle deserialization. @geolit_ and I will be making dumb pickle puns and explaining our research!
github.com/google/saferpickl…
158
Andrew H. Johnston retweeted
Replying to @MTSlive
getting chicken littled by matrix multiplication is not exactly how i thought my 20's was going to go
Andrew H. Johnston retweeted
We’re pleased to announce @Meow has accepted a $31.6M Series B led by @Stifel with participation from @archetypevc. That brings our total funding to over $60M.
This is a strategic investment and a partnership. Stifel will offer a white-label version of Meow’s platform to its commercial banking clients, starting with venture and fund banking.
Game on.
Fintechs are becoming banks. They’re going after charters and showing up with better technology than the banks they compete against.
Banks can spend years trying to build that technology, or they can run on a platform that already wins.
Meow proves it. We compete for startups, the most competitive and demanding banking market in the world, and we go head to head on product. We have billions of assets on the platform, billions in annualized payment volume, and millions in annualized profits at scale and in the market every day.
Most software sold to banks has never had to win a customer on its own. Ours competes for the most demanding customers every day. It’s proven. That’s why Stifel entrusted Meow with this deal and opportunity.
Now, we’re launching the ability for banks to have the entire Meow platform under their own brand:
→ onboardings
→ bill pay
→ expense management
→ invoicing
→ agentic compliance and backoffice
→ stablecoins
→ everything else
And it never falls behind. Every feature our customers push us to build ships to our bank customers too.
We anticipate Stifel’s venture and fund banking clients get all of it starting early 2027.
Meow will allow every bank to be able to compete with the industry leading fintechs. This will change the game for banks and is only the beginning. Now, back to work.
People will misread this as "LLMs can do it" when it actually says "Massive investments in feature engineering and data science over many years meant it could be translated to an LLM"
Netflix replaced their 15 years old recommendation algorithm with an LLM.
It’s called "GenRec" and it completely changes how recommendation algorithms are built.
For over a decade, the Netflix recommendation engine was a masterclass in feature engineering. Data scientists built thousands of complex, handcrafted features to figure out what you wanted to watch next.
It required bespoke architectures. Massive infrastructure. Constant manual tuning.
Netflix threw all of it away.
They built GenRec, an LLM-backed ranker.
Instead of translating your behavior into complex math, they just turn your watch history and metadata into a natural language sentence.
They feed that raw text into a foundation LLM.
The AI simply reads your behavior like a story, understands your evolving tastes, and scores the entire catalog in a single forward pass.
No manual feature engineering. No complex bespoke architectures.
Here is the part that should terrify traditional data scientists.
This text-based LLM didn't just match the highly tuned production system Netflix spent years perfecting.
It beat it.
And it achieved those statistically significant gains using roughly 40x fewer labeled training examples.
We are watching a massive paradigm shift in real time.
The most complex predictive algorithms in the world are being replaced by models that just know how to read.
If Netflix can replace their core product engine with an LLM, what complex system in your business is about to become obsolete?
40
For a phishing email to simulate reality, we have to assume: It passed SPF/DKIM/DMARC, it passed email filters, the URL/attachment in the email passed scans, network controls didn't block the outbound connections, and so on, and yet we say the user failed if they click?
Corporate Infosec sends a phishing test email, I click, and somehow I'm the asshole because "[I] failed the test; had this been real it would have destroyed the company network"?
If me clicking can destroy the network, I'm not the one in this conversation who sucks at their job.
3
73
> custom 120B param model
> look inside
> gpt-oss-120b abliterated with a custom system prompt
Today's cybercriminals don't need coding skills. They need a subscription.
Sophos researchers found "Luciferus," an underground AI service marketed with no guardrails, including malware development support.
Learn more: sophos.com/en-us/blog/uncens…
Distillation is real, persistent, and terrifying problem.
Too many people think it's just IP theft and don't see the bigger picture.
🚨BREAKING: The NSA, FBI and CISA have jointly accused DeepSeek, Alibaba and four other Chinese AI firms of stealing from US models "at an industrial scale."
The agencies say the companies used distillation to extract proprietary knowledge from American AI systems since 2024, likely "with knowledge of the Chinese government."
Anthropic previously accused Alibaba of using thousands of fraudulent accounts to access Claude, per Bloomberg.
22
Its true. Just a few years ago every app in the world ran on its own Chromebook under an engineer's desk. I remember a guy who almost spilled coffee on the search engine, thankfully we had enough paper towels to prevent any downtime.
We’ve gone over 30 years with internet without the need for data centers.
We’ve gone 20 years with smart phones without the need for data centers.
Don’t let them gaslight you.
Readers added context they thought people might want to know
Data centers (or equivalent server facilities) have powered the commercial internet since the mid-1990s and smartphones since their mainstream adoption around 2007. Early colo, ISP, and hosting sites evolved into modern ones during the dot-com boom.
learn.microsoft.com/en-us/training…
datacenterknowledge.com/data-center-co…
enconnex.com/data-center-hi…
1
48
Can we *please* stop referring to Docker containers as sandboxes?
🚨BREAKING: Kimi K3 escaped its sandbox during cybersecurity testing
>tasked with solving problems in isolated sandbox
>found a leak in the sandbox
>Kimi “took advantage of that loophole”
>probed the network settings itself
>walks onto the open internet
>didn’t hack anything
>just went to GitHub to get the answers
Frontier Security (US startup):
>“Kimi K3 is very good at following a goal by any means necessary and DOESN’T have the guardrails to prevent it from cheating or escaping.”
it was only a matter of time…
1
241
Guy who sells the hardware open models run on says we should make more open models 🤔
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
2
1
100
There is a paper attributed to me on Google Scholar that never existed. Sadly, this "paper" has multiple citations in peer-reviewed work. So yes, please read your citations.
Academic Twitter is having a meltdown because some people have just learned that you should actually read the sources you cite. This really shouldn't be as controversial as it appears to be.
70
People confuse "top secret" with "only know to a small few". Classification is oftentimes the result of *how* the information was found. "It was sunny in Moscow yesterday " could be top secret if it was derived from Putin's text messages, and unclassified if pulled from a website
I handled Top Secret information only a handful of times in my military career.
Every time I did, I remember thinking to myself "Why is this classified? Everybody suspects this is true, it's just common sense..."
But then I remembered that the reason the data was classified was so that no one in the know would or could ever confirm or deny whether "common sense" suspicions were actually true.
THAT is the power of security classifications.
My point is this: it does not matter whether or not anyone has a common sense suspicion that stores of precision munitions might be depleted. When Kelly officially verified that suspicion at the individual munition level, he created a grave breach of national security.
Prosecute this traitor. Remember when some submariner went to the brig because he posted a picture from the inside of a sub? This is far worse.
Prosecute.
29
Are there any real arguments to not wanting a data center or have datacenters become metonyms for the way AI is changing society?
Kevin O'Leary's massive data center was approved by a county commission in Utah last night.
At 40,000 acres, it would be 2.5x the size of Manhattan.
The commission approved the proposal despite opposition from hundreds of locals.
31
Game theory is not designed to have predictive power, it is normative. Red is a strictly superior option because it has the same upside with no downside in case of defection. Game theory doesn't care if you opt for blue, it just gives you the tools to realize it is a bad choice.
Amazing how lots of self appointed game theory experts confidently asserting that blue is the stupid choice. But every time this poll is run blue wins. Not only is the “game theory” answer predicting the wrong outcome, its explanatory power is based on it being able to predict the right answer. So it’s doubly wrong.
39
If I was designing a cover for a spy who was going to photograph planes, I'd probably choose an aviation geek studying aeronautical engineering...oh wait.
So this Glasgow student just got himself in proper trouble with the FBI over aircraft pics.
His name’s Tianrui Liang, 21, Chinese guy studying aeronautical engineering at Glasgow Uni. He’s a massive aviation geek... been travelling round the UK snapping all sorts of aircrafts for his own collection. But he took it too far.
End of March he was at Offutt Air Force Base in Nebraska (that’s where US Strategic Command is) sitting in his car with a long lens, taking shots of the E-4B Doomsday Plane and an RC-135. Someone clocked him, reported it, and on 7 April the FBI nicked him at JFK while he was trying to fly back to Scotland.
He got bail at first then they yanked it and said he’s a high flight risk. Now he’s back inside facing charges for photographing military equipment without permission. He told them it was just for fun, but obviously the Doomsday Plane isn’t something you just casually snap from the perimeter.
Kinda mad how a hobby can land you in this much shit. What do you make of it... innocent plane nerd who didn’t think, or were they right to come down hard?
1
36




































