Shawn Sanderson (@SSGoodVibe) header
Shawn Sanderson (@SSGoodVibe) profile picture

AI | Data Privacy | Globally Inspired | Opinions are my own

Joined February 2015
Photo by Shawn Sanderson (@SSGoodVibe)Photo by Shawn Sanderson (@SSGoodVibe)Photo by Shawn Sanderson (@SSGoodVibe)Photo by Shawn Sanderson (@SSGoodVibe)Photo by Shawn Sanderson (@SSGoodVibe)Photo by Shawn Sanderson (@SSGoodVibe)Photo by Shawn Sanderson (@SSGoodVibe)Photo by Shawn Sanderson (@SSGoodVibe)Photo by Shawn Sanderson (@SSGoodVibe)
Spirit's last valuable asset wasn't a plane. It was its data. A judge OK'd Google's $10M buy of Spirit's emails & Teams chats for AI training, after a fight over scrubbing employee PII. De-identify at the source, before your data is someone else's training set.
137
The Spirit Airlines data sale is a preview of every enterprise data deal coming next. Google is paying $10M for roughly 1.9B emails, 664M Teams items and 91M files. But only after sensitive employee records were carved out, everything else de-identified to ISO/IEC 27559, and an independent party verifies it before delivery. The detail that matters most: Google can walk away if de-identification strips out more than 8% of the asset's value. So the real problem was never "can you remove PII." It's whether you can protect people and still deliver data worth buying. That's what we do at Skyflow. We de-identify sensitive data across multimodal sources at scale. Several of the largest AI training-data companies rely on us, with hundreds of terabytes flowing to the AI labs. The data stays private, auditable and valuable. Every company with a data archive just learned it's an asset. Privacy decides what it's worth. #Deidentification #SpiritAirlines #skyflow
117
In Vegas on Dec 3 (re:Invent week)? 🍻 I'm hosting drinks and networking for people in the data acquisition world: data brokers, buyers, RLHF shops, and AI data labs. We'll talk pricing, provenance, and keeping PII out of training sets (my day job at @Skyflow). @KenLaversin @navidrastegar
1
2
3
158
Shawn Sanderson retweeted
The fastest growing companies build with Skyflow. This is one of the best.
We finished another strong quarter at @doppelhq and I’m proud of how far we’ve come. AI has changed the speed and scale of social engineering, with phishing and impersonation attacks increasingly targeting humans and AI agents. That shift was reflected in demand this quarter. Adoption of our agentic Social Engineering Defense (SED) platform continues to accelerate, and we added a record number of new enterprise customers in Q3. Email Security also became the fastest-growing product in Doppel’s history. Q3 Highlights include: - Expanded Human Risk Management with Risk Signal Ingestion, bringing identity and security signals into employee risk profiles - Announced MUFG Innovation Partners as a strategic investor in Doppel - Announced our partnership with the @49ers - Earned @FortuneMagazine Cyber 60 recognition for the second year in a row - Named to the @FastCompany Best Workplaces for Innovators list and earned @GPTW_US certification - Published new brand impersonation research with @a16z - Surpassed 300 employees as we continued growing globally - Opened a new Toronto office, expanding our engineering and go-to-market footprint in Canada - Our team represented Doppel at over 130 in-person events across the globe As always, we’re grateful to our customers for trusting us to protect their organizations as the threat landscape evolves. None of this progress happens without them, or without the partners, investors, and Doppelgangers pushing us forward every day. There’s a lot more to build. We’re hiring!
1
2
6
671
Healthcare AI's real bottleneck isn't compute. It's sanitization. 🧵 Labs are consuming clinical notes, scribe recordings, call center audio and imaging faster than anyone planned. Almost none of it is usable as-is.
2
4
168
At Skyflow we de-identify text, audio, documents and images in one pipeline. Consistent tokens mean a patient stays the same patient across every file without ever being exposed, and authorized teams can reverse it when policy allows.
1
69
The labs that win in healthcare won't have the most data. They'll make regulated data safe at the speed they consume it. What's slowing your pipeline more: access or sanitization? #HealthcareAI #DataPrivacy
5
Healthcare AI's bottleneck isn't compute. It's sanitization. Labs want clinical notes, scribe audio & imaging faster than ever. Scrub the transcript and the audio still says the patient's name. At Skyflow we fix that: one pipeline to de-identify text, audio & images.
1
2
67
This makes total sense: Datasets are the new asset class. And like any asset, value comes down to clean title: provenance, consent, and no PII hiding in the basement.
Today, we are releasing the first off-the-shelf marketplace for company data, built to give AI labs real operating corpora that have been cleared, curated and catalogued. Until now, data deals have happened in Slack channels, Signal chats and late-night Zoom calls. Now any lab can buy company data the way it buys compute. Introducing Grepped Buyside, join the waitlist now: grepped.ai/buyside
1
2
157
When Toys "R" Us Canada sold its brand, customer names and purchase histories went with it. In insolvencies, customer data is a top asset. Buying or selling it? Skyflow de-identifies the PII first, so the value transfers but the liability doesn't. ca.finance.yahoo.com/news/co…
1
43
Good read from @anshublog. The labs bring the model; SaaS brings the system of record and the workflows. The next question every CISO asks: what customer data does the agent actually get to see?
New: Agentforce Claude Coworker. The AI Labs are in alignment with SaaS companies again? How and why? Let me take a shot at explaining.
1
31
LLMs don't have a delete key. Once PII lands in a training set, prompt or vector store, you can't sanitize it back out. De-identify before ingestion. Keep real values in a vault, let models work on tokens. That's what we're solving at Skyflow. #AI #DataPrivacy #LLM
43
Singapore Airlines is the latest carrier hit by the Amadeus breach, after Cathay, JAL, ANA and Korean Air. "Our systems weren't breached" means little to customers whose passport details leaked. Your data is only as safe as every vendor holding a copy. techradar.com/pro/security/s…
89
One of the other challenges is that the sellers of the data want assurances that their data will be fully sanitized; Labs also want this. You need to pick a partner like Skyflow that has the experience with the Labs in doing this.
data brokering is 10x harder than you think. everyone is faking their numbers collecting the data is one of the easiest parts the hard part is selling to the labs directly dont get fooled by people claiming extreme numbers
32
Shawn Sanderson retweeted
Crypto was never decentralized in its implementation.
Excellent comment on this FT article about the Binance hack ft.com/content/1d4dd7b2-9dee…
3
9
892
Malaysia Airlines says its systems were never compromised. Its passengers' data may have been exposed anyway, via reservation vendor Amadeus. At Skyflow, sensitive data stays in a vault and vendors get tokens. A breach exposes tokens, not passports. airwaysmag.com/new-post/mala…
40
Shawn Sanderson retweeted
Agents break data security. You need a runtime agent data control platform.
5
3
2
21
23,782