Blind Insight

Unlock value from data without sacrificing privacy and security. GDPR/CCPA/EHDS/AI-ACT compliant analytics and data-sharing.

[Chaofan Shou @Fried_rice]

Claude code source code has been leaked via a map file in their npm registry!

Code: https://t.co/jBiMoOzt8G

[Gbogr Benedict @bbgbogr]

Most people don’t realize the real bottleneck in AI isn’t compute it’s data access.

You can scale GPUs infinitely, but the moment sensitive data is involved, everything slows down:
compliance, trust, security reviews.
That’s why approaches like encrypted data processing from @GetBlindInsight are interesting making data usable without exposing it.

We’re exploring a similar direction on the compute side with @jungle_grid :
running AI workloads across distributed nodes without requiring trust in the compute layer.

Feels like the future of AI infra is:
secure data × trustless compute.
Curious to see where this goes.

[Jackie Peters @JackiePeters]

Excited to announce that aggregation functions have shipped on @GetBlindInsight! We now support sum, avg, min, max, and count!

[Blind Insight @GetBlindInsight]

🔐 Meet Blind Insight—a complete solution for working with sensitive data in real-time. Blind Insight's searchable encryption is up to billions of times faster than FHE. Fine-grained access controls streamline compliance and minimize exposure.

#PrivacyTech #CyberSecurity #AI

[Blind Insight @GetBlindInsight]

Blind Insight Founder @JackiePeters sat down with @markjeffrey today to discuss the lessons learned from the recent #Coinbase breach and how organizations can be more proactive about their data stewardship.

#DataProtection #DataSecurity #CoinbaseBreach

[Blind Insight @GetBlindInsight]

Who wants to see what 1MM records looks like?

[Jackie Peters @JackiePeters]

Based on this, here are projections on 1MM records and comparisons vs. #FHE on GPU and CPU.

Will release a public @github repo so you all can try it out at home!

[Jackie Peters @JackiePeters]

Hot take: #FHE set encryption-in-use back 5 years by overhyping and underdelivering, and convinced a lot of smart people that #encrypted #ML 'isn't ready yet.'

Meanwhile, we just trained a #fraud model on 50K encrypted records in 90s. On standard hardware. #Demo in comments👇

[Jackie Peters @JackiePeters]

They said it couldn't be done... we were underestimated and under resourced at every turn. But as they say, necessity is the mother of invention, and I have always loved a good challenge.

So proud of our team for pulling this off, and I had so much fun building the example on top of our platform.

[mrinank @MrinankSharma]

Today is my last day at Anthropic. I resigned.

Here is the letter I shared with my colleagues, explaining my decision.

[Jackie Peters @JackiePeters]

Founder wisdoms from Ram Dass

[Deals Dhamaka @AIWorkflowGuide]

While it is easy to condemn this as a personal failure, we should also look at the incentives created by venture capital. When the market rewards growth at any cost, it invites founders to stretch the truth to stay relevant. Roy Lee is responsible for his choices, but this situation reflects a broader culture that prioritizes optics over operational reality. Can the tech industry build a sustainable model where transparency is valued as much as a high multiple?

[Blind Insight @GetBlindInsight]

This is episode 1 of Will It Run On Encrypted Data?. You bring the use case, we’ll try to build it.

What should we run on encrypted data next? Let us know below! 👇

🔗 Link to full 3-minute demo: https://t.co/bQ5FPCRdkc

[Blind Insight @GetBlindInsight]

🚨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. 🧵👇

[Blind Insight @GetBlindInsight]

To prove it, we ran real-time decisioning in milliseconds and validated against a local dataset of 23,500 plaintext records. 100% agreement. Zero accuracy lost.

[Blind Insight @GetBlindInsight]

Who wants to see what 1MM records looks like?

[Jackie Peters @JackiePeters]

Scales sub-linearly with data volume! This time trained the model (naive bayesian classifier) on six features and one target variable, 100K records. 157.2s. 2x the records, only 73.5% latency increase.

[Blind Insight @GetBlindInsight]

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. 🍿

[Jackie Peters @JackiePeters]

Six features, one target variable, 50K records, 94s. 90.8% accuracy, 1:1 results vs. same model trained on plaintext.

[Blind Insight @GetBlindInsight]

If you've evaluated FHE, found it fell short for your use case, and thought that was the only answer, think again.

[Jackie Peters @JackiePeters]

Hot take: #FHE set encryption-in-use back 5 years by overhyping and underdelivering, and convinced a lot of smart people that #encrypted #ML 'isn't ready yet.'

Meanwhile, we just trained a #fraud model on 50K encrypted records in 90s. On standard hardware. #Demo in comments👇

[Blind Insight @GetBlindInsight]

@grubbbyte @MongoDB https://t.co/ACRphzIg56

For more on Blind Insight!