- Get link
- X
- Other Apps
- Get link
- X
- Other Apps
Securing
Artificial Intelligence (SAI)
INTRODUCTION
The speedy enlargement of Artificial Intelligence into new
industries with new stakeholders, coupled with an evolving risk panorama and a huge
boom in AI, presents hard, demanding situations for protection. The ISG SAI
creates high nice technical requirements to fight these demanding situations.
Artificial Intelligence impacts our lives each day, from
nearby AI structures on mobile telephones suggesting the next phrase in our
sentences to massive manufacturers the usage of AI to improve commercial
techniques. AI has the ability to revolutionize our interactions with the generation,
enhance our first-rate of life and enhance safety – but without high nice
technical requirements and true practices, AI has the capacity to create new
attacks and get worse existing security measures.
The ETSI Industry Specification Group on Securing Artificial
Intelligence (ISG SAI) has a key role to play in improving the security of AI
through manufacturing fantastic technical requirements; the ISG SAI will create
standards to preserve and enhance the security of the latest AI technologies.
ROLE & ACTIVITIES
The SAI develops technical specs and reviews to address
three aspects of artificial intelligence in standards: leadmarketingbusiness
Securing AI from assault: wherein AI is part of a system
that desires safety
Mitigating towards malicious AI: wherein AI is used to
improve and beautify traditional attack vectors or create new attack vectors
Using AI to beautify security features: shielding structures
against attack wherein using AI is a part of the ‘answer’ or is used to improve
and decorate greater conventional countermeasures
The ETSI ISG SAI develops the technical understanding that
acts as a baseline in ensuring that synthetic intelligence is secure.
Stakeholders impacted by the activity of ETSI’s institution encompass give up
customers, producers, operators, and governments. technologycompanians
More info is to be had in "Our work."
STANDARDS
A full listing of associated standards in the public area is
out there through the ISG SAI committee web page.
OUR WORK
The ISG SAI first outputs will center around six key
topics:
·
Problem Statement, with the intention to manual
the work of the institution
·
Threat Ontology for AI, to align terminology
·
Data Supply Chain, targeted on facts troubles
and dangers in for schooling AI
·
Mitigation Strategy, with guidance to mitigate
the impact of AI threats
·
Security checking out of AI
·
Role of hardware in the protection of AI
Read on for greater information about every painting object.
Securing AI problem statement
The first SAI record ETSI GR SAI 004, describes the hassle
of securing AI-based totally structures and solutions, with a focal point on
system getting to know and the demanding situations regarding confidentiality,
integrity, and availability at each level of the gadget mastering lifecycle. It
additionally factors out a number of the broader challenges of AI structures
along with bias, ethics, and potential to be explained. A wide variety of
various assault vectors are outlined, as well as several cases of real-global
use and attacks. The suggestions contained in this file will be used to outline
the scope and timescales for the observe-up work.
AI danger ontology
Currently, there is no commonplace know-how of what
constitutes an attack on AI structures, nor the way it is probably created,
hosted, and propagated. This painting will search to define what is considered
an AI threat and the way it differs from threats to conventional structures.
The AI Threat Ontology specification seeks to align
terminology throughout exceptional stakeholders and a couple of industries to
underpin the destiny paintings of the ISG SAI. This will outline unique phrases
within the context of cyber and bodily security, with a narrative this is
without difficulty accessible. This Threat Ontology will cope with AI as a device,
and each as an attacker and a defender of security.
Data supply chain record
Data is an important issue in the development of AI systems,
both uncooked statistics and information and remarks from different AI
structures and humans within the loop. However, access to appropriate facts is
frequently restrained, causing want to the hotel too much less suitable assets
of facts. Compromising the integrity of facts has been tested to be a viable
assault vector in opposition to an AI device.
This document will summarize the methods presently used to
supply information for education AI, along with a review of existing
initiatives for developing information-sharing protocols and analyze
requirements for making sure integrity inside the shared information, data, and
feedback, in addition to the confidentiality of those.
Mitigation strategy report
This painting object will summarize and analyze present and
capability mitigation in opposition to threats for AI-primarily based systems
and produce suggestions for mitigating in opposition to threats introduced with
the aid of adopting AI into systems. These recommendations will shed mild on
protection baselines of AI-primarily based systems by using mitigating in
opposition to recognized or ability security threats. The recommendations may
even deal with security capabilities, demanding situations, and obstacles when
adopting mitigation for AI-primarily based systems in sure use cases.
Security checking out of AI
This work will pick out techniques and strategies for
security testing of AI-based totally additives and convey a radical gap
evaluation to perceive the restrictions and talents in the protection testing
of AI. The hints for safety trying out of AI and AI-based additives will bear in
mind extraordinary algorithms and deal with relevant threats from AI Threat
Ontology paintings.
Role of hardware
The paintings will identify the role of hardware, both
specialized and general-reason, within the security of AI. This will cope with
the mitigations to be had in hardware to prevent assaults and address the
general requirements on hardware to support SAI (expanding from SAI-004). In
addition, this file will cope with feasible techniques to apply AI for the protection
of hardware. The record will also offer a precis of educational and commercial
revel in hardware protection for AI. In addition, the file will deal with
vulnerabilities or weaknesses introduced through hardware that may enlarge
assault vectors on AI.
FUTURE WORK
Although the phrase changed into coined in the 1950s,
realistic AI systems have handiest without a doubt been applied in current
years, driven through:
Evolution of advanced AI strategies, which includes neural
networks, deep getting to know
Availability of considerable statistics sets to allow strong
training
Advances in excessive performance computing enabling
enormously performing gadgets and the availability of hyperscale performance
thru cloud services
These new strategies and abilities, collectively with the
provision of statistics and compute assets, suggest that AI structures will
best come to be extra widely wide-spread. However, this effects a chain of
demanding situations each vintage and new. See beneath for a listing of future
subjects for the ISG SAI.
·
Data security, integrity, and privateness
·
Training statistics: fine, amount,
confidentiality, and labeling
·
Transferability (re-use of fashions across
obligations and industries)
·
Transparency
·
Explainability (for regulation purposes)
·
Ethics and misuse
·
Bias
·
Unintended effects
- Get link
- X
- Other Apps