Open Navigation

IEEE.org IEEE Xplore Digital Library IEEE Standards IEEE Spectrum More Sites

IEEE Transmitter Facebook Twitter Instagram LinkedIn

Open Search
Open Navigation
Open Search

CHALLENGES AND SOLUTIONS

New AI tools are both targets and tools for attackers. Slide to turn challenges into solutions.

Synthetic Reality

Synthetic Reality

Synthetic Reality

Synthetic Reality

The Challenge

AI can generate ultra-realistic images and videos showing people doing or saying things they’ve never done or said. Such “Deepfakes” can make the spread of disinformation more difficult to detect.

The Solution

"AI algorithms have been trained to recognize patterns and characteristics specific to deepfakes. These algorithms analyze visual and audio features, looking for irregularities or deviations from normal patterns."

Carmelo José Albanez Bastos Filho
IEEE Senior Member

DATA LEAKAGE

DATA LEAKAGE

DATA LEAKAGE

DATA LEAKAGE

The Challenge

The information you enter into a chatbot could be used to train it. That could result in models that reveal sensitive or private information.

The Solution

"For businesses dealing with sensitive data, I think we’ll see more AI used in private or contained environments to avoid data being added into public, global databases."

Qiqi Wang
IEEE Member, IEEE Consumer Technology Society

Data in Disguise

Data in Disguise

Data in Disguise

Data in Disguise

The Challenge

Biased, inaccurate, or false data may be injected into AI models to produce erroneous results and flawed systems. AI poisoning techniques have been used in the real world to retrain “spam” filters.

The Solution

"This form of attack has received increasing attention due to its potential to bypass conventional security filters. It has been countered by advanced filtering, monitoring and human supervision systems."

Raul Colcher
IEEE Life Senior Member, IEEE Computer Society, IEEE Communications Society

Advanced Phishing

Advanced Phishing

Advanced Phishing

Advanced Phishing

The Challenge

Phishing scams involve the use of fraudulent emails to trick people or companies into sending money or sharing sensitive data. AI makes it easier to create even more realistic – but still fraudulent – emails and voicemails.

The Solution

"Organizations should consider investing in advanced threat detection and prevention tools. These technologies leverage machine learning and artificial intelligence to identify and flag potential phishing attempts."

Yale Fox
IEEE Member

Evading Guardrails

Evading Guardrails

Evading Guardrails

Evading Guardrails

The Challenge

AI developers typically create guardrails to prevent their creations from generating offensive or illegal content. Plenty of users try anyway – a technique known as “jailbreaking."

The Solution

"There will always be folks who will make efforts to jailbreak systems. The best way to stop them is to more rigorously develop, and then comprehensively test large language model algorithms."

Rebecca Herold
IEEE Member

AI VS AI

AI VS AI

AI VS AI

AI VS AI

The Challenge

AI is being used to exploit or counteract other AI systems -- through techniques like adversarial AI to fool systems that often have critical safety requirements.

The Solution

"By exposing the model to various attack scenarios during the training process, it becomes more robust and capable of recognizing and rejecting malicious inputs."

Amol Gulhane
IEEE Senior Member, IEEE Computer Society

Close Navigation