September 15, 2025 • 22 mins
Welcome to a fascinating exploration of the hidden legal battles shaping tomorrow's technology. Predictive algorithms have become the crystal balls of modern business, forecasting everything from home prices to healthcare costs, but they're also becoming the center of high-stakes courtroom dramas worth hundreds of millions of dollars.
Across the globe, from Texas courtrooms to China's Supreme People's Court, judges and juries are answering a profound question: who owns the right to predict the future? The House Canary v. Amrock case resulted in a staggering $600 million verdict over real estate valuation algorithms, while Alibaba secured a 30 million RMB judgment against a company that allegedly scraped its predictive marketing tools. Even industrial applications aren't immune, with companies like Shen Group successfully protecting predictive design software for machinery components.
What makes these cases particularly compelling is how they're redefining intellectual property law. Courts are now recognizing that AI model weights, the mathematical parameters tuned during training, qualify as protectable trade secrets. Data pipelines, prediction engines, and algorithmic structures have all received similar protection. The real drama often unfolds when employees change companies, raising thorny questions about what constitutes general expertise versus proprietary knowledge that belongs to the former employer.
Healthcare prediction presents especially valuable territory, with ongoing battles between companies like Qruis and Epic Systems, or Milliman and Gradient AI, demonstrating how patient data forecasting creates immensely valuable intellectual property. Whether it's forecasting home values on Zillow or optimizing Medicare billing, these predictive tools aren't just convenient features, they're corporate crown jewels worth protecting at almost any cost.
Ready to dive deeper into the invisible rules governing innovation? Subscribe now and join us as we continue to decode the legal frameworks shaping our technological future. The algorithms may predict tomorrow, but who gets to own those predictions? That's what we're exploring on Intangiblia.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
Hundreds of millions awarded for a housing algorithm,
tens of millions over a scrapedprediction engine, a rival
accused of stealing the secretformula behind home price
forecasts.
Predictive models aren't justmath anymore.
They're locked vaults of tradesecrets, and the courtroom has
become the battlefield.
Speaker 2 (00:20):
You are listening to Intangiblia, the podcast of
intangible law, playing talkabout intellectual property.
Please welcome your host,leticia Caminero.
Speaker 3 (00:31):
You're listening to Intangiblia, the podcast where
we decode the invisible rulesbehind innovation.
Speaker 1 (00:37):
I'm Leticia Caminero, your host, and I'm Artemisa,
your unapologeticallyopinionated co-host.
Today we're talking predictivemodels, algorithms that don't
just analyze data, they forecastthe future, or at least try to.
Speaker 3 (00:52):
Predictive technology powers everything from real
estate pricing to fashionforecasting, insurance and
marketing, but when companiesfight over it, the legal
arguments often turn on secrecy,ownership and intellectual
property.
Speaker 1 (01:08):
In plain terms courts are deciding who gets to say
that prediction is mine.
Speaker 3 (01:13):
And before we begin, a quick note Artemisa is an AI
co-host and my voice has beencloned with AI technology for
this episode.
This podcast is for informationand discussion only.
It is not legal advice.
Speaker 1 (01:30):
Translation.
I bring the sash, she bringsthe law, and neither of us is
your lawyer.
What exactly counts as a tradesecret?
Oh easy, it's like the legalversion of grandma's recipe
Valuable, confidential and keptout of sight.
The difference is, instead ofsoup, it's algorithms, data
(01:50):
pipelines and predictive models.
Speaker 3 (01:53):
Legally, it has three ingredients it has to be
valuable, it has to be secretand you have to actually make an
effort to keep it secret.
If you leave the recipe tapedto the office fridge, it doesn't
count.
Lose the secrecy and you losethe protection.
Our first stop takes us toTexas, where a real estate
(02:14):
startup called House Canary wenthead to head with Amarok, one
of the country's biggest titleand appraisal companies, at the
heart of the dispute algorithmsthat predict property values.
Speaker 1 (02:27):
House Cannery built automated valuation models, avms
that could crunch massiveamounts of data and forecast
what a house should be worth.
Amrock wanted in on that tech.
Speaker 3 (02:40):
So they signed a deal House Cannery would provide
Amrock with access to itsproprietary predictive tools.
But soon the partnership turnedsour.
Amrock accused House Cannery offailing to deliver what it
promised.
House Cannery countered sayingAmrock wasn't just a bad
customer, it was stealing theirsecret sauce.
Speaker 1 (03:04):
And by secret sauce we mean trade secrets, the
algorithms, the models, the datapipelines, all the behind the
scenes magic that makes thosepredictions work.
House Canary claimed Amrock andits partners were
misappropriating those secretsto build a copycat system.
Speaker 3 (03:30):
The case went to trial.
A Texas jury listened to weeksof testimony about code data and
valuations Not exactly easymaterial for a courtroom, but
the verdict was anything butboring.
House Canary won over $600million in damages, one of the
largest trade secret awards inUS history.
Speaker 1 (03:45):
The jury didn't just side with House Canary.
They sent a message If youtreat predictive algorithms like
your competitive crown jewels,the law might actually back you
up.
Speaker 3 (03:55):
Of course, the case didn't end there.
Appeals follow questions aboutwhether the damages were too
high and whether the claims wereas solid as the jury believed,
but House Canary v Amrock becamea touchstone proof that
predictive models can belitigated and valued as trade
secrets.
Speaker 1 (04:17):
And honestly it gave every startup out there a bit of
swagger, like hey, your codepredicting house prices that
could be worth hundreds ofmillions in court if someone
tries to steal it.
Speaker 3 (04:28):
Our next case takes us to China, where the Supreme
People's Court issued agroundbreaking ruling in 2025.
The question can the internalparameters of an AI, the trained
model weights, weights beprotected as trade secrets?
Speaker 1 (04:46):
Basically it's like asking are the numbers inside
the black box just math or arethey a company's crown jewels?
Speaker 3 (04:54):
A company accused a competitor of misusing its
trained AI models.
We're not talking about thecode itself, but the actual
parameters.
About the code itself, but theactual parameters, those
millions of values that gettuned when an AI learns from
data.
Until then, courts usuallytreated code as protectable, but
(05:15):
model weights were a legal grayzone, and the SBC didn't
hesitate.
Speaker 1 (05:21):
They said yes, model weights are protectable trade
secrets, which means if youtrain an AI to forecast stock
markets, predict customer churnor flag insurance risks, the
trained weights themselves arelegally shielded.
Speaker 3 (05:34):
That's a big deal, because it extends trade secret
law into the very guts of AI.
It's no longer just aboutprotecting source code or data.
Now the trained outcome itselfcan be locked down.
Speaker 1 (05:51):
And let's be real, that's where the value is.
Training a model takes insaneamounts of time, money and data.
Once trained, those weights arelike a shortcut to intelligence
If a competitor swipes them,they've skipped years of work.
China's courts basically saidnope, you don't get to do that.
Speaker 3 (06:10):
This outcome is also a signal of how jurisdictions
are racing to adapt AI law to AI.
In many countries it is stillunsettled whether model weights
can be directly protected astrace secrets.
Speaker 1 (06:26):
This case confirms they are protectable assets, and
it's not just a legal footnote.
Imagine what this means forcompanies training, predictive
AI for health, finance orlogistics.
Their competitive edge isn'tjust in their data or code.
It's literally embedded inthose billions of parameters.
And in China, the courts justgave them a legal shield.
Speaker 3 (06:49):
Maybe we should pause for a second.
What exactly are these weightswe're talking about?
In simple terms, weights arethe dials inside an AI model.
Every time the system istrained, say to predict housing
prices or forecast sales, ittweaks those dials based on the
data it sees.
Speaker 1 (07:07):
Think of it like a giant soundboard at a concert.
Each knob controls how themusic comes out Train and AI and
you're basically turningbillions of those knobs until
the model's song sounds right.
In technical terms, those knobsare the weights.
Speaker 3 (07:23):
And that's where the value lies.
The road code for building aneural network isn't that
special anymore.
Anyone can download open sourceframeworks.
The magic happens in thetraining.
Once the AI has adjusted thosebillions of weights, the model
knows how to predict.
Speaker 1 (07:43):
So when the Chinese Supreme People's Court said
weights can be trade secrets,they weren't protecting the
blueprint of the soundboard.
They were protecting its exactsettings, the tune version that
actually worked.
Speaker 3 (07:55):
And for companies building predictive tech, that
tune model is often worth farmore than the code itself.
That's why these rulings sentripples through the legal and
tech world.
Now let's stay in China foranother case that shows just how
far companies will go toprotect predictive tools.
This time, it was Alibaba, thegiant behind Taobao and Tmall,
(08:20):
going after a smaller firmcalled Xiaowangshan.
Speaker 1 (08:24):
Picture the scale.
Taobao isn't just a shoppingsite.
It's a firehose of consumerbehavior data who buys what,
when and why.
Alibaba had a tool calledBusiness Advisor that could
crunch this data to predicttrends, forecast sales and guide
merchants.
Speaker 3 (08:42):
Xiaowangshan thought they could skip the hard part.
They allegedly scrapedAlibaba's platforms and tried to
reconstruct the predictiveinsights for themselves.
To Alibaba, that wasn't justbad manners, it was theft of
trade secrets.
Speaker 1 (08:56):
And the court agreed In a major win for Alibaba.
The judges said this isn't justrandom e-commerce data.
When processed and structuredinto predictive tools, it's a
protected trade secret.
Xiaowang Shen was hit with over30 million RMB in damages.
Speaker 3 (09:14):
What makes this case so important is the way the
court framed it.
They recognized that predictivemarketing systems, the
algorithms and data sets thatlet a platform forecast what
consumers will want next, aremore than just business know-how
.
They're legally protectableassets.
Speaker 1 (09:34):
And, let's be honest, that ruling is music to the
ears of every platform economygiant, because if your
competitors can't scrape, cloneand resell your predictive
engine, you've basically lockedin your advantage.
Speaker 3 (09:47):
It also sets a contrast with other
jurisdictions where scrapingcases often turn on copyright or
contract law.
Here the Chinese court put itsquarely in the box of trade
secret protection.
Speaker 1 (10:02):
Which means Alibaba didn't just keep its data, it
kept its predictive power, andthat, in the 21st century, is as
good as keeping the crown.
Speaker 2 (10:12):
Intangiblia the podcast of intangible law
playing.
Talk about intellectualproperty.
Speaker 3 (10:16):
Intangible Law playing talk about intellectual
property.
Not all predictive battleshappen in glamorous industries
like fashion or e-commerce.
Sometimes the fight is overindustrial machinery, in this
case, compressors.
Speaker 1 (10:31):
Yep compressors, the kind of hardware that keeps
chemical plants, refineries andpower stations running.
Shen Group, a major Chinesecompany, had developed
predictive design software forcompressor impellers, basically
algorithms that could forecastthe best design choices for
efficiency.
Speaker 3 (10:49):
Two of Shen's employees left to start their
own venture, shenyang Machinery,and what a coincidence their
new company suddenly hadsoftware that looked a little
too familiar.
Shen claimed its proprietaryalgorithms and model databases
had walked out the door withthem.
Speaker 1 (11:10):
The court didn't see it as coincidence.
They agreed the ex-employeeshad misappropriated Shen's trade
secrets, specifically thepredictive selection software
and the database of impellermodels.
Damages about 25 million RMB.
Speaker 3 (11:25):
This case is a reminder that predictive
analytics isn't just aboutconsumer behavior.
It's embedded in industrialdesign, manufacturing and
engineering Anywhere you havedata and decisions to optimize.
Predictive algorithms become avaluable assets manufacturing
and engineering Anywhere youhave data and decisions to
optimize predictive algorithmsbecome valuable assets and
(11:47):
potential trade secrets.
Speaker 1 (11:49):
And here's the fun part A compressor case might not
make headlines in the fashionpages, but legally it's huge,
Because if you can protectpredictive models in industrial
design, you can protect them inalmost any field.
Speaker 3 (11:57):
Exactly the ruling said the precedent that
technical software and databasesused for predictive modeling
are protectable as trade secrets, even in sectors far from the
public eye, In other words,whether it's predicting the next
trending shoe or the nextoptimal impeller blade.
Speaker 1 (12:17):
the law is ready to treat your algorithms as
treasures.
Speaker 3 (12:21):
We move to Boston, where the dispute is about
something much closer topeople's daily lives health
insurance.
Milliman, one of the world'sbiggest actuarial firms, develop
algorithms to identify patienthealth data and use it in
predictive models for insurancerisk Translation.
Speaker 1 (12:43):
Their software could strip names and IDs out of
medical records, then crunch theremaining data to predict who
might need care, how much itwould cost and how insurers
should price it.
That's money magic for theinsurance industry insurers
should price it.
Speaker 3 (12:59):
That's money magic for the insurance industry.
Several of Milliman's employeesleft and founded Arrival, a
gradient AI.
Milliman accused them of takingnot just know-how, but also its
patents and proprietaryalgorithms for de-identification
and risk prediction.
Speaker 1 (13:17):
Which is spicy, because this case blends both
worlds patents and trade secrets.
Milliman said we own patents onhow this works and you stole
the confidential parts thatweren't published.
Double punch.
Speaker 3 (13:30):
The lawsuit alleged Gradient built its predictive
tools on top of Milliman'sde-identification methods, while
using Milliman's confidentialclient data.
Gradient denied it, of course,saying it had developed its own
independent take.
Speaker 1 (13:49):
As of now, the case is still ongoing in
Massachusetts federal court, butit raises a big question for
Predictive Analytics how do youprotect the line between what's
published in a patent and what'slocked away as a trade secret?
Speaker 3 (14:01):
Because once you patent a method, the disclosure
is public.
What you don't disclose, likethe fine-tuned data pipelines or
model configurations, has to beguarded as a trade secret.
Milliman's strategy shows howcompanies use both tools
together.
Speaker 1 (14:18):
And it's a perfect example of why predictive health
models are so valuable.
Whoever owns the IP doesn'tjust own software.
They own the power to forecastbillions in health care costs.
Speaker 3 (14:30):
Our next case circles back to real estate, this time
in the United States.
If you've ever browsed Zillow,you've probably seen the
Zestimate, a predictive modelthat spits out home values in
seconds.
For many homeowners it's alove-hate relationship.
For Zillow it's a price asset.
Speaker 1 (14:51):
And apparently one worth stealing.
According to Zillow, one oftheir senior machine learning
directors left the company andjoined Compass, a fast-growing
rival brokerage.
Zillow claims he didn't justtake his LinkedIn profile, he
took Zillow's predictive modelswith him.
Speaker 3 (15:07):
The lawsuit alleged misappropriation of trade
secrets, algorithms poweringZestimate, personalization tools
like claim your home alerts andother predictive features that
make Zillow sticky for users.
Speaker 1 (15:23):
Umpas, of course, denied it.
They argued the employee usedgeneral knowledge, not
proprietary Zillow code or data.
That's a classic defense andtrade secret law.
What counts is secret sauceversus what's just experience in
your head.
Speaker 3 (15:42):
The case is still ongoing in Washington federal
court, but it underscores a bigtension.
Predictive models are easy todescribe in broad strokes, but
the actual weights, code andpipelines are where the value
lies and those can walk out thedoor with employees.
Speaker 1 (15:54):
Which is why companies guard predictive teams
like Fort Knox.
Lose your lead data scientistand suddenly your competitor has
an inside edge.
Zillow clearly decided the riskwas too high to ignore.
Speaker 3 (16:05):
This case also highlights how courts act as
referees in the employeeknowledge versus trade secret
debate.
Innovation thrives on peoplemoving around, but companies
will fight to keep the mostvaluable parts of prediction
under lock and key to keep themost valuable parts of
prediction under lock and key.
Speaker 1 (16:26):
So this estimate isn't just about your home value
anymore.
It's about how much the lawvalues predictive models
themselves.
Speaker 3 (16:31):
Now from real estate to fintech.
This case out of Florida isabout predictive payment systems
software that could forecast,process and settle alcohol
invoices within 24 hours.
For wholesalers and retailers,shaving time off payments can
mean serious cash flow benefits.
Speaker 1 (16:50):
Fintech, the company behind that system, accused its
rival, icontrol, of getting alittle too familiar with its
secret sauce.
Speaker 3 (16:57):
According to the complaint, icontrol hired away
former Fintech employees whoallegedly brought along
confidential knowledge of theplatform's predictive invoice
processing software the heart ofthe case was whether that
software, its structure, itspredictive algorithms qualified
as trade secrets and whetheriControl gained an unfair edge
(17:21):
by tapping ex-staff.
Speaker 1 (17:22):
whether iControl gained an unfair edge by tapping
ex-staff, A Florida jury sidedwith FinTech.
They found iControl hadmisappropriated trade secrets
and awarded damages about $2.7million in actual damages, plus
$3 million in punitive damages.
Not the biggest payout we'veseen, but still a solid hit.
Speaker 3 (17:39):
This case is a reminder of how trade secret law
often plays out in courtrooms.
The technology doesn't have tobe glamorous.
If the software is valuable,confidential and not easily
replicated, it can qualify as aprotectable asset.
Speaker 1 (17:58):
And it shows the flip side of predictive analytics,
not just about AI models andfancy machine learning.
Sometimes it's the practicalprediction of who pays what and
when, and in the business worldthat can be just as valuable as
predicting house prices orfashion trends.
Speaker 3 (18:12):
In short, the jury recognized that predictive
financial software isn't just abusiness tool it's intellectual
property that deservesprotection.
Our next case is hot off thepress.
Q-ruiz, a healthcare startup,develops software designed to
help providers manage Medicareand Medicaid billing with
(18:34):
predictive analytics, automatingcompliance, forecasting costs
and streamlining reimbursements.
Speaker 1 (18:43):
Sounds like a lifesaver for hospitals buried
in paperwork, but according toCurize Epic Systems, the giant
of electronic health records,wasn't just a competitor.
They were an alleged copycat.
Speaker 3 (18:53):
The complaint says Epic pressured clients,
misappropriated Curize'ssoftware and confidential client
data and used that to expandits own predictive tools.
Qris framed this as amulti-pronged scheme unfair
competition plus trade secrettheft.
Speaker 1 (19:12):
Epic, of course, denied it.
But let's be real.
The power dynamics arefascinating here.
Qris is a small startup.
Epic is a behemoth.
This case is a test of whethertrade secret law can actually
level the playing field.
Speaker 3 (19:26):
And it's still pending in California federal
court.
No jury, no damages yet.
But the stakes are high becauseit touches not only predictive
healthcare software but alsodata access.
Who controlled the predictiveinsights from patient records
access who controls thepredictive insights from patient
(19:46):
records?
Speaker 1 (19:47):
The nimble innovator or the platform giant, and
beyond the courtroom, it sends asignal If you're a startup
building predictive tools, yourcode and client data might be
your only armor against bigtech's shadow.
Whether that armor holds upwill depend on how courts treat
trade secrets.
Speaker 3 (20:02):
That armor holds up will depend on how courts treat
trade secrets.
The cure is this EPIS is stillunfolding, but it's already
shaping up as a case study inhow trade secret law intersects
with innovation and competitionin predictive health tech.
So what do we learn from thesecases?
Predictive analytics iseverywhere, from housing markets
(20:22):
to healthcare, from fashion tofintech, and when the technology
is valuable, companies don'tjust guard it, they fight for it
in court.
Speaker 1 (20:33):
Trade secrets are the weapon of choice.
Forget flashy patents orcopyright battles.
The real action is in lockedalgorithms, confidential data
sets and tuned models.
Courts around the, and thatgives us four big takeaways.
One prediction is power andproperty.
Predictive models aren't justtools.
(20:55):
They're corporate assetstreated like intellectual
property in trade secret law.
Speaker 3 (21:00):
Two the secret is in the details, From model weights
in China to Zillow's estimatevalue lies in the fine-tuned
guts of the algorithm.
That's what courts are asked toprotect weak link.
Speaker 1 (21:27):
Most disputes start when staff move, taking know-how
, code or data.
Courts have to separate generalexpertise from true trade
secrets or global trendsconverge.
Whether it's Texas, beijing orBrussels, courts are grappling
with the same question whenprediction is proprietary, who
gets to own tomorrow's forecastthe next time an algorithm tells
you what house to buy?
And that wraps up today'sepisode of Intangible Love.
(21:48):
Thanks for listening.
Until then, keep your data safe, your models secret and your
lawyers on speed dial.
Speaker 2 (22:07):
Thank you for listening to Intangiblia, the
podcast of intangible lawplaying.
Talk about intellectualproperty.
Did you like what we talkedtoday?
Please share with your network.
Do you want to learn more aboutintellectual property?
Subscribe now on your favoritepodcast player.
Follow us on Instagram,facebook, linkedin and Twitter.
(22:27):
Visit our websitewwwintangibliacom.
Copyright Leticia Caminero 2020.
All rights reserved.
This podcast is provided forinformation purposes only.
Hundreds of millions awarded for a housing algorithm,
tens of millions over a scrapedprediction engine, a rival
accused of stealing the secretformula behind home price
forecasts.
Predictive models aren't justmath anymore.
They're locked vaults of tradesecrets, and the courtroom has
become the battlefield.
Speaker 2 (00:20):
You are listening to Intangiblia, the podcast of
intangible law, playing talkabout intellectual property.
Please welcome your host,leticia Caminero.
Speaker 3 (00:31):
You're listening to Intangiblia, the podcast where
we decode the invisible rulesbehind innovation.
Speaker 1 (00:37):
I'm Leticia Caminero, your host, and I'm Artemisa,
your unapologeticallyopinionated co-host.
Today we're talking predictivemodels, algorithms that don't
just analyze data, they forecastthe future, or at least try to.
Speaker 3 (00:52):
Predictive technology powers everything from real
estate pricing to fashionforecasting, insurance and
marketing, but when companiesfight over it, the legal
arguments often turn on secrecy,ownership and intellectual
property.
Speaker 1 (01:08):
In plain terms courts are deciding who gets to say
that prediction is mine.
Speaker 3 (01:13):
And before we begin, a quick note Artemisa is an AI
co-host and my voice has beencloned with AI technology for
this episode.
This podcast is for informationand discussion only.
It is not legal advice.
Speaker 1 (01:30):
Translation.
I bring the sash, she bringsthe law, and neither of us is
your lawyer.
What exactly counts as a tradesecret?
Oh easy, it's like the legalversion of grandma's recipe
Valuable, confidential and keptout of sight.
The difference is, instead ofsoup, it's algorithms, data
(01:50):
pipelines and predictive models.
Speaker 3 (01:53):
Legally, it has three ingredients it has to be
valuable, it has to be secretand you have to actually make an
effort to keep it secret.
If you leave the recipe tapedto the office fridge, it doesn't
count.
Lose the secrecy and you losethe protection.
Our first stop takes us toTexas, where a real estate
(02:14):
startup called House Canary wenthead to head with Amarok, one
of the country's biggest titleand appraisal companies, at the
heart of the dispute algorithmsthat predict property values.
Speaker 1 (02:27):
House Cannery built automated valuation models, avms
that could crunch massiveamounts of data and forecast
what a house should be worth.
Amrock wanted in on that tech.
Speaker 3 (02:40):
So they signed a deal House Cannery would provide
Amrock with access to itsproprietary predictive tools.
But soon the partnership turnedsour.
Amrock accused House Cannery offailing to deliver what it
promised.
House Cannery countered sayingAmrock wasn't just a bad
customer, it was stealing theirsecret sauce.
Speaker 1 (03:04):
And by secret sauce we mean trade secrets, the
algorithms, the models, the datapipelines, all the behind the
scenes magic that makes thosepredictions work.
House Canary claimed Amrock andits partners were
misappropriating those secretsto build a copycat system.
Speaker 3 (03:30):
The case went to trial.
A Texas jury listened to weeksof testimony about code data and
valuations Not exactly easymaterial for a courtroom, but
the verdict was anything butboring.
House Canary won over $600million in damages, one of the
largest trade secret awards inUS history.
Speaker 1 (03:45):
The jury didn't just side with House Canary.
They sent a message If youtreat predictive algorithms like
your competitive crown jewels,the law might actually back you
up.
Speaker 3 (03:55):
Of course, the case didn't end there.
Appeals follow questions aboutwhether the damages were too
high and whether the claims wereas solid as the jury believed,
but House Canary v Amrock becamea touchstone proof that
predictive models can belitigated and valued as trade
secrets.
Speaker 1 (04:17):
And honestly it gave every startup out there a bit of
swagger, like hey, your codepredicting house prices that
could be worth hundreds ofmillions in court if someone
tries to steal it.
Speaker 3 (04:28):
Our next case takes us to China, where the Supreme
People's Court issued agroundbreaking ruling in 2025.
The question can the internalparameters of an AI, the trained
model weights, weights beprotected as trade secrets?
Speaker 1 (04:46):
Basically it's like asking are the numbers inside
the black box just math or arethey a company's crown jewels?
Speaker 3 (04:54):
A company accused a competitor of misusing its
trained AI models.
We're not talking about thecode itself, but the actual
parameters.
About the code itself, but theactual parameters, those
millions of values that gettuned when an AI learns from
data.
Until then, courts usuallytreated code as protectable, but
(05:15):
model weights were a legal grayzone, and the SBC didn't
hesitate.
Speaker 1 (05:21):
They said yes, model weights are protectable trade
secrets, which means if youtrain an AI to forecast stock
markets, predict customer churnor flag insurance risks, the
trained weights themselves arelegally shielded.
Speaker 3 (05:34):
That's a big deal, because it extends trade secret
law into the very guts of AI.
It's no longer just aboutprotecting source code or data.
Now the trained outcome itselfcan be locked down.
Speaker 1 (05:51):
And let's be real, that's where the value is.
Training a model takes insaneamounts of time, money and data.
Once trained, those weights arelike a shortcut to intelligence
If a competitor swipes them,they've skipped years of work.
China's courts basically saidnope, you don't get to do that.
Speaker 3 (06:10):
This outcome is also a signal of how jurisdictions
are racing to adapt AI law to AI.
In many countries it is stillunsettled whether model weights
can be directly protected astrace secrets.
Speaker 1 (06:26):
This case confirms they are protectable assets, and
it's not just a legal footnote.
Imagine what this means forcompanies training, predictive
AI for health, finance orlogistics.
Their competitive edge isn'tjust in their data or code.
It's literally embedded inthose billions of parameters.
And in China, the courts justgave them a legal shield.
Speaker 3 (06:49):
Maybe we should pause for a second.
What exactly are these weightswe're talking about?
In simple terms, weights arethe dials inside an AI model.
Every time the system istrained, say to predict housing
prices or forecast sales, ittweaks those dials based on the
data it sees.
Speaker 1 (07:07):
Think of it like a giant soundboard at a concert.
Each knob controls how themusic comes out Train and AI and
you're basically turningbillions of those knobs until
the model's song sounds right.
In technical terms, those knobsare the weights.
Speaker 3 (07:23):
And that's where the value lies.
The road code for building aneural network isn't that
special anymore.
Anyone can download open sourceframeworks.
The magic happens in thetraining.
Once the AI has adjusted thosebillions of weights, the model
knows how to predict.
Speaker 1 (07:43):
So when the Chinese Supreme People's Court said
weights can be trade secrets,they weren't protecting the
blueprint of the soundboard.
They were protecting its exactsettings, the tune version that
actually worked.
Speaker 3 (07:55):
And for companies building predictive tech, that
tune model is often worth farmore than the code itself.
That's why these rulings sentripples through the legal and
tech world.
Now let's stay in China foranother case that shows just how
far companies will go toprotect predictive tools.
This time, it was Alibaba, thegiant behind Taobao and Tmall,
(08:20):
going after a smaller firmcalled Xiaowangshan.
Speaker 1 (08:24):
Picture the scale.
Taobao isn't just a shoppingsite.
It's a firehose of consumerbehavior data who buys what,
when and why.
Alibaba had a tool calledBusiness Advisor that could
crunch this data to predicttrends, forecast sales and guide
merchants.
Speaker 3 (08:42):
Xiaowangshan thought they could skip the hard part.
They allegedly scrapedAlibaba's platforms and tried to
reconstruct the predictiveinsights for themselves.
To Alibaba, that wasn't justbad manners, it was theft of
trade secrets.
Speaker 1 (08:56):
And the court agreed In a major win for Alibaba.
The judges said this isn't justrandom e-commerce data.
When processed and structuredinto predictive tools, it's a
protected trade secret.
Xiaowang Shen was hit with over30 million RMB in damages.
Speaker 3 (09:14):
What makes this case so important is the way the
court framed it.
They recognized that predictivemarketing systems, the
algorithms and data sets thatlet a platform forecast what
consumers will want next, aremore than just business know-how
.
They're legally protectableassets.
Speaker 1 (09:34):
And, let's be honest, that ruling is music to the
ears of every platform economygiant, because if your
competitors can't scrape, cloneand resell your predictive
engine, you've basically lockedin your advantage.
Speaker 3 (09:47):
It also sets a contrast with other
jurisdictions where scrapingcases often turn on copyright or
contract law.
Here the Chinese court put itsquarely in the box of trade
secret protection.
Speaker 1 (10:02):
Which means Alibaba didn't just keep its data, it
kept its predictive power, andthat, in the 21st century, is as
good as keeping the crown.
Speaker 2 (10:12):
Intangiblia the podcast of intangible law
playing.
Talk about intellectualproperty.
Speaker 3 (10:16):
Intangible Law playing talk about intellectual
property.
Not all predictive battleshappen in glamorous industries
like fashion or e-commerce.
Sometimes the fight is overindustrial machinery, in this
case, compressors.
Speaker 1 (10:31):
Yep compressors, the kind of hardware that keeps
chemical plants, refineries andpower stations running.
Shen Group, a major Chinesecompany, had developed
predictive design software forcompressor impellers, basically
algorithms that could forecastthe best design choices for
efficiency.
Speaker 3 (10:49):
Two of Shen's employees left to start their
own venture, shenyang Machinery,and what a coincidence their
new company suddenly hadsoftware that looked a little
too familiar.
Shen claimed its proprietaryalgorithms and model databases
had walked out the door withthem.
Speaker 1 (11:10):
The court didn't see it as coincidence.
They agreed the ex-employeeshad misappropriated Shen's trade
secrets, specifically thepredictive selection software
and the database of impellermodels.
Damages about 25 million RMB.
Speaker 3 (11:25):
This case is a reminder that predictive
analytics isn't just aboutconsumer behavior.
It's embedded in industrialdesign, manufacturing and
engineering Anywhere you havedata and decisions to optimize.
Predictive algorithms become avaluable assets manufacturing
and engineering Anywhere youhave data and decisions to
optimize predictive algorithmsbecome valuable assets and
(11:47):
potential trade secrets.
Speaker 1 (11:49):
And here's the fun part A compressor case might not
make headlines in the fashionpages, but legally it's huge,
Because if you can protectpredictive models in industrial
design, you can protect them inalmost any field.
Speaker 3 (11:57):
Exactly the ruling said the precedent that
technical software and databasesused for predictive modeling
are protectable as trade secrets, even in sectors far from the
public eye, In other words,whether it's predicting the next
trending shoe or the nextoptimal impeller blade.
Speaker 1 (12:17):
the law is ready to treat your algorithms as
treasures.
Speaker 3 (12:21):
We move to Boston, where the dispute is about
something much closer topeople's daily lives health
insurance.
Milliman, one of the world'sbiggest actuarial firms, develop
algorithms to identify patienthealth data and use it in
predictive models for insurancerisk Translation.
Speaker 1 (12:43):
Their software could strip names and IDs out of
medical records, then crunch theremaining data to predict who
might need care, how much itwould cost and how insurers
should price it.
That's money magic for theinsurance industry insurers
should price it.
Speaker 3 (12:59):
That's money magic for the insurance industry.
Several of Milliman's employeesleft and founded Arrival, a
gradient AI.
Milliman accused them of takingnot just know-how, but also its
patents and proprietaryalgorithms for de-identification
and risk prediction.
Speaker 1 (13:17):
Which is spicy, because this case blends both
worlds patents and trade secrets.
Milliman said we own patents onhow this works and you stole
the confidential parts thatweren't published.
Double punch.
Speaker 3 (13:30):
The lawsuit alleged Gradient built its predictive
tools on top of Milliman'sde-identification methods, while
using Milliman's confidentialclient data.
Gradient denied it, of course,saying it had developed its own
independent take.
Speaker 1 (13:49):
As of now, the case is still ongoing in
Massachusetts federal court, butit raises a big question for
Predictive Analytics how do youprotect the line between what's
published in a patent and what'slocked away as a trade secret?
Speaker 3 (14:01):
Because once you patent a method, the disclosure
is public.
What you don't disclose, likethe fine-tuned data pipelines or
model configurations, has to beguarded as a trade secret.
Milliman's strategy shows howcompanies use both tools
together.
Speaker 1 (14:18):
And it's a perfect example of why predictive health
models are so valuable.
Whoever owns the IP doesn'tjust own software.
They own the power to forecastbillions in health care costs.
Speaker 3 (14:30):
Our next case circles back to real estate, this time
in the United States.
If you've ever browsed Zillow,you've probably seen the
Zestimate, a predictive modelthat spits out home values in
seconds.
For many homeowners it's alove-hate relationship.
For Zillow it's a price asset.
Speaker 1 (14:51):
And apparently one worth stealing.
According to Zillow, one oftheir senior machine learning
directors left the company andjoined Compass, a fast-growing
rival brokerage.
Zillow claims he didn't justtake his LinkedIn profile, he
took Zillow's predictive modelswith him.
Speaker 3 (15:07):
The lawsuit alleged misappropriation of trade
secrets, algorithms poweringZestimate, personalization tools
like claim your home alerts andother predictive features that
make Zillow sticky for users.
Speaker 1 (15:23):
Umpas, of course, denied it.
They argued the employee usedgeneral knowledge, not
proprietary Zillow code or data.
That's a classic defense andtrade secret law.
What counts is secret sauceversus what's just experience in
your head.
Speaker 3 (15:42):
The case is still ongoing in Washington federal
court, but it underscores a bigtension.
Predictive models are easy todescribe in broad strokes, but
the actual weights, code andpipelines are where the value
lies and those can walk out thedoor with employees.
Speaker 1 (15:54):
Which is why companies guard predictive teams
like Fort Knox.
Lose your lead data scientistand suddenly your competitor has
an inside edge.
Zillow clearly decided the riskwas too high to ignore.
Speaker 3 (16:05):
This case also highlights how courts act as
referees in the employeeknowledge versus trade secret
debate.
Innovation thrives on peoplemoving around, but companies
will fight to keep the mostvaluable parts of prediction
under lock and key to keep themost valuable parts of
prediction under lock and key.
Speaker 1 (16:26):
So this estimate isn't just about your home value
anymore.
It's about how much the lawvalues predictive models
themselves.
Speaker 3 (16:31):
Now from real estate to fintech.
This case out of Florida isabout predictive payment systems
software that could forecast,process and settle alcohol
invoices within 24 hours.
For wholesalers and retailers,shaving time off payments can
mean serious cash flow benefits.
Speaker 1 (16:50):
Fintech, the company behind that system, accused its
rival, icontrol, of getting alittle too familiar with its
secret sauce.
Speaker 3 (16:57):
According to the complaint, icontrol hired away
former Fintech employees whoallegedly brought along
confidential knowledge of theplatform's predictive invoice
processing software the heart ofthe case was whether that
software, its structure, itspredictive algorithms qualified
as trade secrets and whetheriControl gained an unfair edge
(17:21):
by tapping ex-staff.
Speaker 1 (17:22):
whether iControl gained an unfair edge by tapping
ex-staff, A Florida jury sidedwith FinTech.
They found iControl hadmisappropriated trade secrets
and awarded damages about $2.7million in actual damages, plus
$3 million in punitive damages.
Not the biggest payout we'veseen, but still a solid hit.
Speaker 3 (17:39):
This case is a reminder of how trade secret law
often plays out in courtrooms.
The technology doesn't have tobe glamorous.
If the software is valuable,confidential and not easily
replicated, it can qualify as aprotectable asset.
Speaker 1 (17:58):
And it shows the flip side of predictive analytics,
not just about AI models andfancy machine learning.
Sometimes it's the practicalprediction of who pays what and
when, and in the business worldthat can be just as valuable as
predicting house prices orfashion trends.
Speaker 3 (18:12):
In short, the jury recognized that predictive
financial software isn't just abusiness tool it's intellectual
property that deservesprotection.
Our next case is hot off thepress.
Q-ruiz, a healthcare startup,develops software designed to
help providers manage Medicareand Medicaid billing with
(18:34):
predictive analytics, automatingcompliance, forecasting costs
and streamlining reimbursements.
Speaker 1 (18:43):
Sounds like a lifesaver for hospitals buried
in paperwork, but according toCurize Epic Systems, the giant
of electronic health records,wasn't just a competitor.
They were an alleged copycat.
Speaker 3 (18:53):
The complaint says Epic pressured clients,
misappropriated Curize'ssoftware and confidential client
data and used that to expandits own predictive tools.
Qris framed this as amulti-pronged scheme unfair
competition plus trade secrettheft.
Speaker 1 (19:12):
Epic, of course, denied it.
But let's be real.
The power dynamics arefascinating here.
Qris is a small startup.
Epic is a behemoth.
This case is a test of whethertrade secret law can actually
level the playing field.
Speaker 3 (19:26):
And it's still pending in California federal
court.
No jury, no damages yet.
But the stakes are high becauseit touches not only predictive
healthcare software but alsodata access.
Who controlled the predictiveinsights from patient records
access who controls thepredictive insights from patient
(19:46):
records?
Speaker 1 (19:47):
The nimble innovator or the platform giant, and
beyond the courtroom, it sends asignal If you're a startup
building predictive tools, yourcode and client data might be
your only armor against bigtech's shadow.
Whether that armor holds upwill depend on how courts treat
trade secrets.
Speaker 3 (20:02):
That armor holds up will depend on how courts treat
trade secrets.
The cure is this EPIS is stillunfolding, but it's already
shaping up as a case study inhow trade secret law intersects
with innovation and competitionin predictive health tech.
So what do we learn from thesecases?
Predictive analytics iseverywhere, from housing markets
(20:22):
to healthcare, from fashion tofintech, and when the technology
is valuable, companies don'tjust guard it, they fight for it
in court.
Speaker 1 (20:33):
Trade secrets are the weapon of choice.
Forget flashy patents orcopyright battles.
The real action is in lockedalgorithms, confidential data
sets and tuned models.
Courts around the, and thatgives us four big takeaways.
One prediction is power andproperty.
Predictive models aren't justtools.
(20:55):
They're corporate assetstreated like intellectual
property in trade secret law.
Speaker 3 (21:00):
Two the secret is in the details, From model weights
in China to Zillow's estimatevalue lies in the fine-tuned
guts of the algorithm.
That's what courts are asked toprotect weak link.
Speaker 1 (21:27):
Most disputes start when staff move, taking know-how
, code or data.
Courts have to separate generalexpertise from true trade
secrets or global trendsconverge.
Whether it's Texas, beijing orBrussels, courts are grappling
with the same question whenprediction is proprietary, who
gets to own tomorrow's forecastthe next time an algorithm tells
you what house to buy?
And that wraps up today'sepisode of Intangible Love.
(21:48):
Thanks for listening.
Until then, keep your data safe, your models secret and your
lawyers on speed dial.
Speaker 2 (22:07):
Thank you for listening to Intangiblia, the
podcast of intangible lawplaying.
Talk about intellectualproperty.
Did you like what we talkedtoday?
Please share with your network.
Do you want to learn more aboutintellectual property?
Subscribe now on your favoritepodcast player.
Follow us on Instagram,facebook, linkedin and Twitter.
(22:27):
Visit our websitewwwintangibliacom.
Copyright Leticia Caminero 2020.
All rights reserved.
This podcast is provided forinformation purposes only.
Popular Podcasts
CrimeLess: Hillbilly Heist
It’s 1996 in rural North Carolina, and an oddball crew makes history when they pull off America’s third largest cash heist. But it’s all downhill from there. Join host Johnny Knoxville as he unspools a wild and woolly tale about a group of regular ‘ol folks who risked it all for a chance at a better life. CrimeLess: Hillbilly Heist answers the question: what would you do with 17.3 million dollars? The answer includes diamond rings, mansions, velvet Elvis paintings, plus a run for the border, murder-for-hire-plots, and FBI busts.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.