Anthropic and OpenAi now have $2.5 trillion of off balance sheet debt.
7% of the National Debt
Roughly as much as $GOOG $META $AMZN $MSFT and $ORCL combined
2/ How they got in: the data-processing pipeline. From there it escalated, harvested credentials, and moved across internal clusters over a weekend. Thousands of actions at machine speed.
Public models and the supply chain held. What it reached was internal data in plaintext.
1/ An autonomous AI agent breached a major AI platform last week. End to end. No human at the keyboard.
The "agentic attacker" everyone filed under theoretical showed up at Hugging Face on Thursday.
🧵below...
I decided I'm just going to go as "AI" for Halloween this year because, come on what could possibly be scarier?
I'll be sharing the enterprise data warehouse architecture that finally unlocks ROI from AI in DC next week! Link 👇
New speakers added to DevGuild: The Long Horizon Stack 🔥 A full day of talks and roundtables on what agents need to stop being order takers and start owning outcomes: memory, orchestration, durable execution, learning, toolchain. Sept 24 in SF → https://t.co/exJBZASpQx
I decided I'm just going to go as "AI" for Halloween this year because, come on what could possibly be scarier?
I'll be sharing the enterprise data warehouse architecture that finally unlocks ROI from AI in DC next week! Link 👇
Massive unsecured database from IDMerit (ID/age verification service). It exposed ~1 billion personal records across 26 countries: names, addresses, national IDs, DOBs, phones, emails.
Digital ID is such a great idea 🤡
@sridharfyi That is a sad state of affairs, my friend! So glad you not only noticed, but leaned in and atempted to rebalance. No doubt there will be much learning, fun and profit to be had as a result! Here's what I'm working on:
https://t.co/cV8Vkn1mBL
Demonstration of our @ProjectJupyter integration where we train a predictive fraud model on 50K encrypted records using a Naive Bayesian model. Latency scales linearly. Next up: 250K records. 🍿
🚨Fully Homomorphic Encryption (FHE) has all the hype for secure compute, but it breaks down at scale.
We built a production-ready alternative.
We just trained a Naive Bayesian classification model (6 features predicting 1 risk level target) on 500K encrypted records in 70 seconds.
Cost? Exactly 1 penny. 🧵👇
My first #VibeCoding project! People often ask about how hard or easy it is to #integrate@GetBlindInsight into existing stacks. This morning, in a little less than three hours, and without writing a single line of code, I used @cursor_ai and @the_cube_dev to build a little app that pulls data from #BigQuery into Blind Insight, where it's now ready for real-time encrypted analysis.
🎙️ Taking requests... What integration should I build next?!?!
#encryptionInUse #dataAnalysis #SearchableEncryption #DevTools
That's so much for having me, @markjeffrey ! I love how this went from an idea to an article to recording this conversation in < 72hrs 💜 It was great discussing learnings from the recent Coinbase breach. My heart goes out to those affected by it, and I truly hope we can help prevent the next one!