AI security is not only about protecting a model after deployment. Compromised training data can ...

AI security is not only about protecting a model after deployment. Compromised training data can ...AI security is not only about protecting a model after deployment. Compromised training data can ...

The network for creativity

Join 1.25M professional creatives like you

Connect with clients, get discovered, and run your business 100% commission-free

Creatives on Contra have earned over $150M and we are just getting started

Back to feedPost

Stephen Kisong'e

• Aug 5

AI security is not only about protecting a model after deployment.

Compromised training data can cause a network intrusion model to learn the wrong patterns, miss threats, and create a false sense of security.

Businesses using AI should ask:

• Can the training data be trusted? • Who can modify the data pipeline? • Has the model been tested against manipulation? • Are false negatives reviewed before deployment?

Check it out

Bash Hugging Face Python

Adversarial Data Poisoning Attack on a Network Intrusion AI Mod…

pro

• Aug 17

Case study is out! 🚀

Wrapped up a very interesting mobile project in just 4 weeks from ideas and exploration to a polished final product.

The full case study is now live. Would love to hear your thoughts!

Check it out

Claude Figma Jitter

Victrex Graphics

• Aug 18

Awesome!

Révolté

pro

• Aug 17

Some brands get moodboarded. This one got excavated.

Pampa case study is live — a data brand where the logic of a 2,000-year-old desert marking and the vein structure of a butterfly wing turned out to be the same idea wearing two different ages.

Check it out

Brand Design Creative Direction Logo Design

Asif Aleem

pro

• Aug 18

The strongest part is that you found one underlying logic that reads across a 2,000-year-old marking and a butterfly wing, so the mark feels discovered rather than styled. That is the line between a moodboard and a system: the concept still holds when you change the medium.