Darktrace announces ISO/IEC 42001 accreditation

Darktrace is thrilled to announce that we are one of the first cybersecurity companies to achieve ISO/IEC 42001 accreditation for the responsible management of AI systems.

This is a critical milestone for Darktrace as we continue to strengthen our offering, mature our governance and compliance frameworks for AI management, expand our research and development capabilities, and further our commitment to the development of responsible AI.  

It cements our commitment to providing secure, trustworthy and proactive cybersecurity solutions that our customers can rely on and complements our existing compliance framework, consisting of certifications for:

• ISO/IEC 27001:2022 – Information Security Management System
• ISO/IEC 27018:2019 – Protection of Personally Identifiable Information in Public Cloud Environments
• Cyber Essentials – A UK Government-backed certification scheme for cybersecurity baselines

What is ISO/IEC 42001:2023?

In response to the unique challenges that AI poses, the International Organization for Standardization (ISO) introduced the ISO/IEC 42001:2023 framework in December 2023 to help organizations providing or utilizing AI-based products or services to demonstrate responsible development and use of AI systems. To achieve the accreditation, organizations are required to establish, implement, maintain, and continually improve their Artificial Intelligence Management System (AIMS).

ISO/IEC 42001:2023 is the first of its kind, providing valuable guidance for this rapidly changing field of technology. It addresses the unique ethical and technical challenges AI poses by setting out a structured way to manage risks such as transparency, accuracy and misuse without losing opportunities. By design, it balances the benefits of innovation against the necessity of a proper governance structure.

Being certified means the organization has met the requirements of the ISO/IEC 42001 standard, is conforming to all applicable regulatory and legislative requirements, and has implemented thorough processes to address AI risks and opportunities.

What is the  ISO/IEC 42001:2023 accreditation process?

Darktrace partnered with BSI over an 11-month period to undertake the accreditation. The process involved developing and implementing a comprehensive AI management system that builds on our existing certified frameworks, address the risks and opportunities of  using and developing cutting-edge AI systems, underpins our AI objectives and policies, and meets our regulatory and legal compliance requirements.

The AI Management System, which takes in our people, processes, and products, was extensively audited by BSI against the requirements of the standard, covering all aspects spanning the design of our AI, use of AI within the organization, and our competencies, resources and HR processes. It is an in-depth process that we’re thrilled to have undertaken, making us one of the first in our industry to achieve certification for a globally recognized AI system.

The scope of Darktrace’s certification is particularly wide due to our unique Self-Learning approach to AI for cybersecurity, which uses multi-layered AI systems consisting of varied AI techniques to address distinct cybersecurity tasks. The certification encompasses production and provision of AI systems based on anomaly detection, clustering, classifiers, regressors, neural networks, proprietary and third-party large language models for proactive, detection, response and recovery cybersecurity applications. Darktrace additionally elected to adopt all Annex A controls present in the ISO/IEC 42001 standard.

What are the benefits of an AI Management System?

While AI is not a new or novel concept, the AI industry has accelerated at an unprecedented rate in the past few years, increasing operational efficiency, driving innovation, and automating cumbersome processes in the workplace.

At the same time, the data privacy, security and bias risks created by rapid innovation in AI have been well documented.

Thus, an AI Management System enables organizations to confidently establish and adhere to governance in a way that conforms to best practice, promotes adherence, and is in line with current and emerging regulatory standards.

Not only is this vital in a unique and rapidly evolving field like AI, it additionally helps organization’s balance the drive for innovation with the risks the technology can present, helping to get the best out of their AI development and usage.

What are the key components of ISO/IEC 42001?

The Standard puts an emphasis on responsible AI development and use, requiring organizations to:

• Establish and implement an AI Management System
• Commit to the responsible development of AI against established, measurable objectives
• Have in place a process to manage, monitor and adapt to risks in an effective manner
• Commit to continuous improvement of their AI Management System

The AI Standard is similar in composition to other ISO standards, such as ISO/IEC 27001:2022, which many organizations may already be familiar with. Further information as to the structure of ISO/IEC 42001 can be found in Annex A.

What it means for Darktrace’s customers

Our certification against ISO/IEC 24001 demonstrates Darktrace’s commitment to delivering industry-leading Self-Learning AI in the name of cybersecurity resilience. Our stakeholders, customers and partners can be confident that Darktrace is responsibly, ethically and securely developing its AI systems, and is managing the use of AI in our day-to-day operations in a compliant, secure and ethical manner. It means:

• ‍You can trust our AI: We can demonstrate our AI is developed responsibly, in a transparent manner and in accordance with ethical rules. For more information and to learn about Darktrace's responsible AI in cybersecurity approach, please see here.‍
• Our products are backed by innovation and integrity: Darktrace drives cutting edge AI innovation with ethical governance and customer trust at its core. ‍
• You are partnering with an organization which stays ahead of regulatory changes: In an evolving AI landscape, partnering with Darktrace helps you to stay prepared for emerging compliance and regulatory demands in your supply chain.

Achieving ISO/IEC 42001:2023 certification is not just a checkpoint for us. It represents our unwavering commitment to setting a higher standard for AI in cybersecurity. It reaffirms our leadership in building and implementing responsible AI and underscores our mission to continuously innovate and lead the way in the industry.

Annex A: The Structure of ISO/IEC 42001

ISO/IEC 42001 has requirements for which seven adherence is required for an organization seeking to obtain or maintain its certification:

• ‍Context of the organization – organizations need to demonstrate an understanding of the internal and external factors influencing the organization’s AI Management System.
• ‍Leadership – senior leadership teams need to be committed to implementing AI governance within their organizations, providing direction and support across all aspects AI Management System lifecycle.‍
• Planning – organizations need to put meaningful and manageable processes in place to identify risks and opportunities related to the AI Management System to achieve responsible AI objectives and mitigate identified risks.‍
• Support – demonstrating a commitment to provisioning of adequate resources, information, competencies, awareness and communication for the AI Management System is a must to ensure that proper oversight and management of the system and its risks can be achieved.‍
• Operation – establishing processes necessary to support the organization’s AI system development and usage, in conformance with the organization’s AI policy, objectives and requirements of the standard. Correcting the course of any deviations within good time is paramount.‍
• Performance evaluation – the organization must be able to demonstrate that it has the capability and willingness to regularly monitor and evaluate the performance of the AI Management System effectively, including actioning any corrections and introducing new processes where relevant.‍
• Improvement – relying on an existing process will not be sufficient to ensure compliance with the AI Standard. Organizations must commit to monitoring of existing systems and processes to ensure that the AI Management System is continually enhanced and improved.

To assist organizations in seeking the above, four annexes are included within the AI Standard’s rubric, which outline the objectives and measures an organization may wish to implement to address risks related to the design and operation of their AI Management System through the introduction of normative controls. Whilst they are not prescriptive, Darktrace has implemented the requirements of these Annexes to enable it to appropriately demonstrate the effectiveness of its AI Management System. We have placed a heavy emphasis on Annex A which contains these normative controls which we, and other organizations seeking to achieve certification, can align with to address the objectives and measures, such as:

• Enforcement of policies related to AI.
• Setting responsibilities within the organization, and expectation of roles and responsibilities.
• Creating processes and guidelines for escalating and handling AI concerns.
• Making resources for AI systems available to users.
• Assessing impacts of AI systems internally and externally.
• Implementing processes across the entire AI system life cycle.
• Understanding treatment of Data for AI systems.
• Defining what information is, and should be available, for AI systems.
• Considering and defining use cases for the AI systems.
• Considering the impact of the AI System on third-party and customer relationships.

The remaining annexes provide guidance on implementing Annex A’s controls, objectives and primary risk sources of AI implementation, and considering how the AI Management System can be used across domains or sectors responsibly.

[related-resource]

Most cloud environments are over-permissioned and under-prepared for incident response.

Security teams need access to logs, snapshots, and configuration data to understand how an attack unfolded, but giving blanket access opens the door to insider threats, misconfigurations, and lateral movement.

So, how do you enable forensics without compromising your security posture?

The dilemma: balancing access and security

There is a tension between two crucial aspects of cloud security that create a challenge for cloud forensics.

One aspect is the need for Security Operations Center (SOC) and Incident Response (IR) teams to access comprehensive data for investigating and resolving security incidents.

The other conflicting aspect is the principle of least privilege and minimal manual access advocated by cloud security best practices.

This conflict is particularly pronounced in modern cloud environments, where traditional physical access controls no longer apply, and infrastructure-as-code and containerization have transformed the landscape.

There are several common but less-than-ideal approaches to this challenge:

• Accepting limited data access, potentially leaving incidents unresolved
• Granting root-level access during major incidents, risking further compromise

Relying on cloud or DevOps teams to retrieve data, causing delays and potential miscommunication

[related-resource]

Challenges in container forensics

Containers present unique challenges for forensic investigations due to their ephemeral and dynamic nature. The orchestration and management of containers, whether on private clusters or using services like AWS Elastic Kubernetes Service (EKS), introduce complexities in capturing and analyzing forensic data.

To effectively investigate containers, it's often necessary to acquire the underlying volume of a node or perform memory captures. However, these actions require specific Identity and Access Management (IAM) and network access to the node, as well as familiarity with the container environment, which may not always be straightforward.

An alternative method of collection in containerized environments is to utilize automated tools to collect this evidence. Since they can detect malicious activity and collect relevant data without needing human input, they can act immediately, securing evidence that might be lost by the time a human analyst is available to collect it manually.

Additionally, automation can help significantly with access and permissions. Instead of analysts needing the correct permissions for the account, service, and node, as well as deep knowledge of the container service itself, for any container from which they wish to collect logs. They can instead collect them, and have them all presented in one place, at the click of a button.

A better approach: practical strategies for cloud forensics

It's crucial to implement strategies that strike a balance between necessary access and stringent security controls.

Here are several key approaches:

1. Dedicated cloud forensics accounts

Establishing a separate cloud account or subscription specifically for forensic activities is foundational. This approach isolates forensic activities from regular operations, preventing potential contamination from compromised environments. Dedicated accounts also enable tighter control over access policies, ensuring that forensic operations do not inadvertently expose sensitive data to unauthorized users.

A separate account allows for:

• Isolation: The forensic investigation environment is isolated from potentially compromised environments, reducing the risk of cross-contamination.
• Tighter access controls: Policies and controls can be more strictly enforced in a dedicated account, reducing the likelihood of unauthorized access.
• Simplified governance: A clear and simplified chain of custody for digital evidence is easier to maintain, ensuring that forensic activities meet legal and regulatory requirements.

For more specifics:

• AWS strategies
• Azure strategies
• GCP strategies

2. Cross-account roles with least privilege

Using cross-account IAM roles, the forensics account can access other accounts, but only with permissions that are strictly necessary for the investigation. This ensures that the principle of least privilege is upheld, reducing the risk of unauthorized access or data exposure during the forensic process.

3. Temporary credentials for just-in-time access

Leveraging temporary credentials, such as AWS STS tokens, allows for just-in-time access during an investigation. These credentials are short-lived and scoped to specific resources, ensuring that access is granted only when absolutely necessary and is automatically revoked after the investigation is completed. This reduces the window of opportunity for potential attackers to exploit elevated permissions.

For AWS, you can use commands such as:

aws sts get-session-token --duration-seconds 43200

aws sts assume-role --role-arn role-to-assume --role-session-name "sts-session-1" --duration-seconds 43200

For Azure, you can use commands such as:

az ad app credential reset --id <appId> --password <sp_password> --end-date 2024-01-01

For more details for Google Cloud environments, see “Create short-lived credentials for a service account” and the request.time parameter.

4. Tag-based access control

Pre-deploying access control based on resource tags is another effective strategy. By tagging resources with identifiers like "Forensics," access can be dynamically granted only to those resources that are relevant to the investigation. This targeted approach minimizes the risk of overexposure and ensures that forensic teams can quickly and efficiently access the data they need.

For example, in AWS:

Condition: StringLike: aws:ResourceTag/Name: ForensicsEnabled

Condition: StringLike: ssm:resourceTag/SSMEnabled: True

For example, in Azure:

"Condition": "StringLike(Resource[Microsoft.Resources/tags.example_key], '*')"

For example, in Google Cloud:

expression: > resource.matchTag('tagKeys/ForensicsEnabled', '*')

Tighten access, enhance security

The shift to cloud environments demands a rethinking of how we approach forensic investigations. By implementing strategies like dedicated cloud forensic accounts, cross-account roles, temporary credentials, and tag-based access control, organizations can strike the right balance between access and security. These practices not only enhance the effectiveness of forensic investigations but also ensure that access is tightly controlled, reducing the risk of exacerbating an incident or compromising the investigation.

Find the right tools for your cloud security

Darktrace delivers a proactive approach to cyber resilience in a single cybersecurity platform, including cloud coverage.

Darktrace’s cloud offerings have been bolstered with the acquisition of Cado Security Ltd., which enables security teams to gain immediate access to forensic-level data in multi-cloud, container, serverless, SaaS, and on-premises environments.

In addition to having these forensics capabilities, Darktrace / CLOUD is a real-time Cloud Detection and Response (CDR) solution built with advanced AI to make cloud security accessible to all security teams and SOCs. By using multiple machine learning techniques, Darktrace brings unprecedented visibility, threat detection, investigation, and incident response to hybrid and multi-cloud environments.