Okay, We Know AI Is Coming. But What About Its Security Risks?

You're building a cool AI (Artificial Intelligence) startup.

Suddenly, your AI system fails.

Or worse, it gets hacked in to.

What went wrong?

What is AI Security?

AI is becoming really powerful.

And there's an increasing dependence on it.

But companies have stated security and privacy as their top barrier to implementing it.

In order to protect your AI, it's important to know the risks.

Types of AI Risks

1. AI Algorithm Testing:

Testing AI is hard because of how complex and expensive it can be. And when AI is not tested properly, it fails during production. For example, some teams only measure a single aggregate metric instead of looking at all the subsets. Companies like Robust Intelligence help test and measure AI systems at scale.

2. AI Hacking/Attacking

‍Imagine people hacking into your self-driving car or your Alexa. These new AI attacks are aimed at the underlying machine learning algorithms. Examples include Evasion Attacks (disrupting the model) and Poisoning Attacks (contaminate the data set).

Companies to Watch:

Counterfit - Microsoft's open source tool to test attacking your own AI

Robust Intelligence: AI model monitoring

Protopia.ai - data privacy for AI

Fiddler - AI continuous monitoring

Protegrity - data protection for AI

3 Startup Opportunities:

1. AI security as a service: Become the expert to help companies with AI security best practices. You can help test their algorithms and protect them against attacks.
2. AI security audits: Create a standard for AI security audits. You test their algorithms and if they pass the audit, they are certified to deploy.
3. AI continuous monitoring: After an AI is deployed, it's important to continuously evaluate the algorithm as things change often. Build a platform that can do this.

‍