AI for security Flashcards

1
Q

How do AI and cyber security relate?

A

AI for cybersecurity: AI is used to improve defensive cybersecurity

Malicious AI: Used to enhance offensive cybersecurity and malicious abuse of AI to manipulate capabilities of AI systems

Cybersecurity for AI: Cybersecurity is used to protect AI systems and users

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

What is an impotant research direction covering all 3 dimensions?

A

Protecting AI systems and users from malicous AI using AI

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

What is the purpose of AI for cybersecurity?

A

Improving security solutions and predict future attacker’s behaviour

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Name an example of AI-driven cybersecurity approaches

A

Automated cyber defense and cyber threat intelligence operations for prevention, analysis, detection and response to cyber threats and incidents

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What are the benefits of the AI-driven security approaches?

A

Less manual effort and less time consuming

Better cope with increasingly and interconnected modern environments

Learn weak signals unnoticed by humans

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What is the 5 stages of NIST syber cecurity framework?

A

Identify
Protect
Detect
Respond
Recover

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Where is anomaly detection used?

A

Across domains where identifying unusual patterns or deviations from the norm is crucial

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

What is the purpose of malicious AI?

A

Expanding the cyber threat landscape

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

What are the benefit of the approach of malicious AI?

A

Sophistication - more targeted
Speed - Automated
Scale

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Give an example of malicious abuse of AI

A

Adversarial machine learning

Trying to subvert existing AI systems to alter their capabilities

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

What can an attacker do to a generic ML based system (7)

A

Insert sensor errors

Modify input image (digital attack)

Alter model weight (model poisoning attacks)

interfere outputs (Output attacks)

Hack learning algorithm (algorithm poisoning attack)

Poison training data (data poisoning attack)

Modify scene (physical attacks)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Define AI-based cyber attacks

A

The application of AI-driven techniques in the attack process, which can be used in conjunction with conventional attack techniques to cause greater damage

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

What is malicious use of AI

A

AI-based cyber attacks

Improving attacker’s capabilities

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Name some use cases for AI-based attacks (5)

A

next-generation malware (highly targeted, evasive)

Voice synthesis (imitating someone’s voice)

password based attacks (learn password distribution)

Social bots (tailored phishing messages)

Adversarial training (automated generation of adversarial examples)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

What is the purpose of cybersecurity for AI

A

Protecting AI systems and users against different types of threats

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

What are the 2 main approaches on securing AI?

A

A narrow and traditional scope: The protection of AI systems against attacks across the AI lifecycle

A broad and extended scope: Holistic view on protecting AI systems through trustworthiness features

17
Q

What are some benefits of securing AI?

A

Contribute to a more secure, safe and fair design, development and operation of AI systems

More robust solutions when using AI for cyber security

18
Q

What are some challenges on securing AI?

A
  1. Revisiting threat models of AI systems from both other AIs and conventional methods
  2. Considering AI systems in the real world
  3. Assessing the human factor
  4. Concept drift: AI systems may evolve over time, thus security properties may degrade.
  5. Interactions and trade offs between security and other AI trustworthiness principles