🔗 https://catarisklab.com
🔗 linkedin.com/in/anthonycata
🔗 huggingface.co/Cata-Risk-Lab
🔗 github.com/dcata004
the ai act applies to any system processing eu resident data. doesn't matter where you're incorporated.
audited a uk saas with 40% eu users. zero documentation. penalties up to 7% of global turnover.
check your user geography.
the ai act applies to any system processing eu resident data. doesn't matter where you're incorporated.
audited a uk saas with 40% eu users. zero documentation. penalties up to 7% of global turnover.
check your user geography.
We built Veritas—a Judge LLM that catches hallucinations before your compliance team doesn't.
Open source. Because "trust me" isn't an audit trail.
#InfoSec #OpenSource #AI
We built Veritas—a Judge LLM that catches hallucinations before your compliance team doesn't.
Open source. Because "trust me" isn't an audit trail.
#InfoSec #OpenSource #AI
the ciso had no idea.
built a tool to map this.
the ciso had no idea.
built a tool to map this.
So we fixed it: £4.5k for due diligence. £12.5k for the full governance suite. Fixed price. No "Phase 2" upsells. No hourly billing surprises.
Just the data you need. Clean, fast, done.
So we fixed it: £4.5k for due diligence. £12.5k for the full governance suite. Fixed price. No "Phase 2" upsells. No hourly billing surprises.
Just the data you need. Clean, fast, done.
Does your AI touch data from someone in Munich? Cool. You're now subject to the EU AI Act. €35M fines don't care where your HQ is.
Geography is dead. Jurisdiction is everything.
#EUAIAct
Does your AI touch data from someone in Munich? Cool. You're now subject to the EU AI Act. €35M fines don't care where your HQ is.
Geography is dead. Jurisdiction is everything.
#EUAIAct
what works: one page. red/amber/green. that's it.
red = stop immediately
amber = fix within 30 days
green = proceed
released a sanitized template. link in reply.
what works: one page. red/amber/green. that's it.
red = stop immediately
amber = fix within 30 days
green = proceed
released a sanitized template. link in reply.
exposure it addressed: £400k+
m&a deal that didn't stall in due diligence: £2.8m
governance isn't a cost center. it's the cheapest insurance you can buy.
exposure it addressed: £400k+
m&a deal that didn't stall in due diligence: £2.8m
governance isn't a cost center. it's the cheapest insurance you can buy.
eu ai act wants risk classification.
australian soci act wants forensic proof of data residency.
one policy document cannot satisfy three incompatible frameworks. you need a jurisdictional heatmap, not a generic compliance binder.
eu ai act wants risk classification.
australian soci act wants forensic proof of data residency.
one policy document cannot satisfy three incompatible frameworks. you need a jurisdictional heatmap, not a generic compliance binder.
released our audit tools as open source:
- wattle-guard (australian soci/app 8)
- swiss risk calculator (nfadp/eu ai act)
- veritas (rag hallucination auditor)
repos in reply. use them. fork them. improve them.
released our audit tools as open source:
- wattle-guard (australian soci/app 8)
- swiss risk calculator (nfadp/eu ai act)
- veritas (rag hallucination auditor)
repos in reply. use them. fork them. improve them.
You can't just run an AI credit scorer. You have to explain its logic to the customer. In writing. On demand.
One firm just ate CHF 250k because their vendor was a black box.
If you can't explain the sausage, don't serve it.
You can't just run an AI credit scorer. You have to explain its logic to the customer. In writing. On demand.
One firm just ate CHF 250k because their vendor was a black box.
If you can't explain the sausage, don't serve it.
built wattle-guard repo in reply.
built wattle-guard repo in reply.
built veritas to fix this. it runs a judge protocol against your rag system, flags every claim that can't trace back to a source doc.
turns "we think it works" into "here's the quantified error rate."
repo in reply.
built veritas to fix this. it runs a judge protocol against your rag system, flags every claim that can't trace back to a source doc.
turns "we think it works" into "here's the quantified error rate."
repo in reply.
a geneva firm's credit scoring ai was accurate and profitable. but when a rejected applicant asked "why?" they couldn't answer.
penalty. ai offline. still.
a geneva firm's credit scoring ai was accurate and profitable. but when a rejected applicant asked "why?" they couldn't answer.
penalty. ai offline. still.