AI Access Security

Securing access to GenAI applications

Presenter Name | Title

Date
 AI Adoption is Transforming Workforce Productivity
Conversational Chats

Native AI apps are exploding

Code Assistants & Generators

Video & Image Generators 12,000+

Projected AI apps in use by 2030.

Writing Assistants Source: Pitchbook's Generative AI Emerging Space, Artificial Intelligence &
Machine Learning Analyst Curated Vertical and SaaS Vertical

And more…

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 As GenAI App Usage Increases,
So Do the Security Risks

Shadow AI apps
create security blind spots. IOTW: Samsung employees allegedly leak proprietary
information via ChatGPT
Three separate employees have allegedly leaked information to the AI chatbot

55%
of employees have
Sensitive data loss
stems from uninspected GenAI
US House forbids staff members from using AI chatbot
Microsoft Copilot
House Office of Cybersecurity has deemed Microsoft Copilot a risk to users because of the
prompts and responses. threat of leaking House data to non-House approved cloud services.

used unapproved
GenAI tools at work.

Malicious content Microsoft accidentally exposes 38TB of internal data via

GitHub repository
from GenAI responses 38 terabytes of internal data, including passwords, publicly accessible through a
GitHub repository that Microsoft’s artificial intelligence research group uses
poses risk to users.

Source: Salesforce 2023

© 2024 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 GenAI Apps Need Rigorous
ChatGPT4
Enter your message to start a chat. This could be a question, a statement, or any other text. Scrutiny for Data Uploads
You Employees may unknowingly share
E Can you help me optimize the following code? confidential data when using GenAI apps.
import boto3 Uploaded data can be used to train AI
aws_access_key_id=‘ACCESS_KEY_ID_HERE’
aws_access_key_id=‘SECRET_KEY_ID_HERE’
models, potentially compromising proprietary
aws_session_token=‘SESSION_TOKEN_HERE’#Optional, only needed if you are information or leading to unintended data leak.
using temporary credentials

#Create a session using your credentials

session-botos3.Session(
aws_access_key_id=aws_access_key_id,
aws_secres_access_key=aws_secret_access_key,
aws_session_token=aws_session_token#Remove if not using temporary credentials)

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 AI Meeting Assistant Notes Search Record

Undetected GenAI Plugins

Home
My Conversations
Weekly Team Meeting Edit May Lead to Unauthorized
Ralph Duncan
Data Access
Groups
Good Morning. Welcome to this week’s team morning. First off, I’d like to
General congratulate everyone. All of our efforts and hard work has paid off. We reached our
Marketing goal and sold $10 million the last quarter of 2024. Congratulations! Interconnected SaaS ecosystems with
Product design
Ralph Duncan
AI-powered third-party integrations create
Sales
Support We also closed a big deal in Canada last week. Everything’s looking good. security vulnerabilities, complicating
Hyun-Jae, how’s our client in Japan? detection and control.
Direct Messages
Su Hyun-Jae
Steven
Charlies
We’re experiencing little hiccups Ralph. The last time I talked to the Marketing This leads to increased risk of security
Team they are coming up with solutions on how to solve this problem. They are
James
looking into rebrand our Japan client. But I have to follow them up on that. breaches and non-compliance with
Folders
Mitchel should get the details about an upcoming launch this Thursday. regulatory requirements.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 What’s Needed to Secure GenAI App Adoption
Challenges

Democratization of AI app AI apps store, learn, and AI apps ingest Rapidly evolving, powerful
creation has led to sprawl reiterate data. unstructured inputs and apps are readily accessible
of insecure apps. generate diverse outputs. through marketplaces.

PRODUCTIVITY

GenAI APP

CODE ASSISTANTS/GENERATORS

RISK HIGH

ENCRYPTION NO
WRITING ASSISTANTS PASSWORDS CREDIT CARD
MODEL TRAINING YES
SSN SECRETS

Comprehensive GenAI Visibility into AI apps Granular data controls with Visibility and control of
app catalog that keeps that train on data. context-aware ML based detectors. 3rd-party AI plugins.
pace with the boom.
Ability to protect against threats Detect plugins with
in GenAI responses. excessive permissions.

Solutions

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 Introducing AI Access Security
to Enable Safe AI Adoption

Real-time visibility of AI usage

View what AI apps are used and by whom.

Access control
Block unsanctioned apps, apply infosec
policies, and protect against threats.

Comprehensive data protection

Scan what data, secrets, and IP are shared.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 Visibility App Access Controls Data Access Controls Security Controls Continuous Monitoring

Step 1 Visibility Into AI Apps

Discover GenAI app usage across

different use cases.

Detailed catalog of 1K+ GenAI

applications.

In-depth visibility into 60+ application

attributes.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 Visibility App Access Controls Data Access Controls Security Controls Continuous Monitoring

Step 2 Classification and App Access Controls

Classify applications across

sanctioned, tolerated, and unsanctioned.

Get tailored visibility and control

measures for each use case.

Set up robust application access controls

with OOTB best practice policies.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 Visibility App Access Controls Data Access Controls Security Controls Continuous Monitoring

Step 3 Data Access Controls

Utilize LLM powered and context-aware

ML models to classify data across
300+ categories.

Set up contextual inline policies to

prevent sensitive data exfiltration to
GenAI apps.

End-user coaching via Prisma Access

agent and browser integrations.

Visibility into encrypted traffic directly

through Prisma Access Browser.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 Visibility App Access Controls Data Access Controls Security Controls Continuous Monitoring

Step 4 Security Controls

Uncover interconnected GenAI apps

within SaaS marketplaces.

Identify, monitor, and remediate

unauthorized AI bots.

Detect threats (malicious URLs, files)

within GenAI app responses.

Monitor and maintain GenAI app

posture for compliance.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 Visibility App Access Controls Data Access Controls Security Controls Continuous Monitoring

Step 5 Continuous Risk Monitoring

Review app adoption and usage insights

across GenAI app categories.

Comprehensive report on GenAI app

usage, risks, security, and compliance.

Contextual recommendations to
strengthen security controls for GenAI apps.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
Demo

© 2024 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 AI Access Security Customer Stories
Multinational high-tech Multinational Multinational Large financial firm
company consulting firm retail company
Safely enable GenAI and prevent Full visibility and access to monitor Prevent sensitive data exposure Discover GenAI and Define AI
source code exposure all GenAI apps in use via GenAI apps data governance

● All GenAI apps classified, 240 allowed ● Discovered 397 GenAI apps accessed ● 92 GenAI apps were accessed by ● 76 GenAI apps were not detected by
(60 Sanctioned, 180 Tolerated) by employees regularly employees manual efforts to block GenAI
● 3 risky apps discovered and blocked ● 3 high risk GenAI apps were discovered ● 16 Threats were detected in first ● 21 threats were detected and
with Zero Trust Access Control in use and blocked 30 days stopped in first 30 days
● 1,531 incidents of Sensitive data leaks ● 643 threats were detected from GenAI ● 12 incidents of sensitive data leaks ● 9 attempts of sensitive data leaks
blocked in first 30 days apps in first 30 days were detected in first 30 days were prevented in first 30 days

“With a fully defined and configured our “We’ve gained visibility into all GenAI “Seeing all GenAI in use helped us shape “We discovered all GenAI apps on our
GenAI Governance policy framework, we applications on our network. Employees our strategy. Tracking threats and data network and cut admin time from 30
confidently embrace generative AI while can safely use almost any GenAI tool while leaks in these tools made refining our AI hours a week to less than one. With a
ensuring data security, compliance, and we seamlessly restrict only the most governance easy and effective.” clear governance framework, we can
responsible usage—without adding high-risk applications.” InfoSec Director now control risky apps and prevent
operational complexity.” Security Director sensitive data leaks.”
CISO CISO

© 2024 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 How to Get Started With AI Access Security

Learn more about Schedule a customized

our capabilities. product demo.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
Thank You

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.
 CASE STUDY

Palo Alto Networks IT

TEAM ● Discover all GenAI apps on the network.

SOUGHT TO… “While AI applications
● Categorize existing and emerging GenAI apps. increase the productivity for
our employees, they can also
● Prevent sensitive data loss.
introduce security issues. AI
● Coach end users on safe GenAI use. Access Security enables my
security teams to have not
only comprehensive visibility
into AI apps but also access
KEY IMPACTS ● Discovered 124 GenAI apps with three new apps weekly. control and continuous
● Classified apps as Sanctioned (11), Tolerated (18),
monitoring.”
and Unsanctioned (95) with the ability to automatically
Meerah Rajavel
block new apps. CIO, Palo Alto Networks

● Implemented 12 new DLP policies to protect sensitive

source code, secrets, access tokens, and credentials.

© 2024
2023 Palo Alto Networks, Inc. All rights reserved. Proprietary and confidential information.