L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
AI Risk for Business Leaders · Module 5 · Lesson 1

The AI Vendor Landscape: Dependency, Opacity, and Concentration

When you buy an AI product, you inherit every risk your vendor took to build it.

On March 20, 2023, OpenAI disclosed that a bug in the Redis open-source library it used had caused approximately 1.2% of ChatGPT Plus subscribers to briefly see another active user's chat titles, first messages, and in some cases partial payment information. The vulnerability was not in OpenAI's own code — it was in a third-party caching layer OpenAI had integrated. Hundreds of thousands of businesses that had embedded ChatGPT into customer workflows suddenly faced questions about data exposure they had no direct visibility into and no contractual warning of.

The incident illustrated a principle that governs every modern AI deployment: your risk surface is not bounded by your own walls. It extends through every vendor, library, and foundation model provider in your stack.

How AI Vendor Dependency Actually Works

Most enterprise AI products are not purpose-built from scratch. They are assembled stacks: a foundation model (GPT-4, Claude, Gemini, Llama) sits at the base; a middleware layer handles orchestration and retrieval; vector databases store embeddings; monitoring tools observe outputs. Each layer is typically provided by a different vendor, often under different contractual terms, different security certifications, and different data-handling regimes.

When your organisation licenses an AI application from a SaaS vendor, you are in practice entering relationships with every subprocessor that vendor uses — yet your procurement process likely evaluated only the top-level vendor. The subprocessor chain is frequently three to five companies deep before you reach the actual model inference infrastructure.

This creates what researchers at the Brookings Institution described in 2023 as "n-tier opacity": the business deploying AI cannot observe the risk practices of the vendors that actually perform inference on its data. Standard SaaS due diligence — SOC 2 reports, penetration test summaries, data processing agreements — typically covers only the immediate vendor, not the chain.

Concentration Risk: The Big-Three Foundation Model Problem

As of 2024, the majority of commercial AI applications globally run on inference infrastructure controlled by three providers: OpenAI, Google (Vertex AI / Gemini), and Anthropic, with Microsoft Azure OpenAI Service acting as a fourth major distribution channel. Amazon Bedrock aggregates several of these behind AWS infrastructure. This means a service disruption, policy change, or security incident at any one of these providers propagates simultaneously across thousands of downstream enterprise applications.

This is not hypothetical. In November 2023, OpenAI experienced a governance crisis — the board's abrupt firing and reinstatement of CEO Sam Altman over five days — that created acute uncertainty for enterprises whose AI roadmaps depended entirely on GPT-4 access. Companies that had not built provider-agnostic architectures had no credible fallback. The episode prompted Goldman Sachs, JPMorgan, and several European banks to formally document "AI provider concentration risk" in internal risk registers for the first time.

The practical implication: dependency on a single foundation model provider is a business continuity risk, not merely a technology preference. Organisations should maintain documented fallback procedures and, where critical workloads are involved, test failover to an alternative provider at least annually.

Key Risk: Model Deprecation Without Notice

In June 2023, OpenAI deprecated the original GPT-3.5 and Codex model endpoints used by thousands of production applications, giving 90 days' notice. Applications that had hardcoded model names — common in rapid prototypes that became production systems — broke at deprecation. Organisations should treat AI model versions as infrastructure with end-of-life dates and conduct periodic "model continuity" audits of all AI integrations.

Key Terms for AI Vendor Risk
Subprocessor ChainThe sequence of third-party companies that process your data downstream of your primary AI vendor. Often two to five layers deep and rarely fully disclosed in standard contracts.
Model Deprecation RiskThe risk that a foundation model version your application depends on is retired, changed, or altered in capability without adequate notice, breaking dependent workflows.
Provider Concentration RiskOver-reliance on a single AI infrastructure provider such that a disruption at that provider propagates as a business continuity event across your AI-dependent operations.
N-Tier OpacityThe inability of an AI deploying organisation to observe, audit, or assess the security and governance practices of vendors beyond the first tier of the supply chain.
Business Leader Takeaway

Every AI vendor relationship is actually a portfolio of vendor relationships you cannot see. Your vendor due diligence process must explicitly demand full subprocessor disclosure, review each subprocessor's data handling terms, and map your concentration exposure across foundation model providers before approving production deployment.

Lesson 1 Quiz

3 questions — free, untracked, retake anytime.
The March 2023 ChatGPT data exposure incident originated in which component?
✓ Correct. The vulnerability was in the open-source Redis library, a third-party dependency — not in OpenAI's own code. This illustrates how supply-chain risks propagate upward to end users.
✗ The incident originated in a third-party Redis caching library OpenAI had integrated, not in OpenAI's own code or authentication systems.
What is "n-tier opacity" in the context of AI vendor risk?
✓ Correct. N-tier opacity means standard due diligence covers only the immediate vendor, leaving the security and governance practices of subprocessors two to five layers deep invisible to the deploying organisation.
✗ N-tier opacity refers to the deploying organisation's inability to see or audit the practices of vendors beyond its direct vendor — the deeper subprocessor chain remains opaque.
Which event in November 2023 caused enterprises to formally document "AI provider concentration risk" in risk registers?
✓ Correct. The five-day governance crisis at OpenAI in November 2023 prompted banks including Goldman Sachs and JPMorgan to add AI provider concentration risk to formal risk registers — highlighting how governance instability at a vendor creates enterprise risk.
✗ The trigger was the OpenAI governance crisis of November 2023, when the board's firing and reinstatement of Sam Altman over five days revealed how dependent enterprises had become on a single provider's continuity.

Lab 1: AI Vendor Dependency Mapping

Practice auditing your AI vendor stack and identifying concentration risks.

Mapping Your AI Subprocessor Chain

In this lab you will practice the skills of AI vendor dependency analysis: identifying subprocessor chains, mapping concentration risk across foundation model providers, and formulating due diligence questions for vendor contracts.

Describe a real or hypothetical AI deployment at your organisation and the assistant will help you map its dependency structure, identify concentration risks, and draft the right questions for your vendor procurement process.

Try asking: "We use a CRM that runs on Salesforce Einstein, which I believe uses OpenAI under the hood. We also use Microsoft Copilot for M365. How do I think about our concentration risk here?" — or describe your own AI stack.
AI Vendor Risk Lab Lab 1
AI Risk for Business Leaders · Module 5 · Lesson 2

Data Flows in AI Supply Chains: What Leaves Your Organisation and Where It Goes

The data you submit to an AI vendor does not stop moving when the response comes back.

In March 2023, Samsung's semiconductor division discovered that engineers had on at least three separate occasions pasted proprietary chip design schematics, internal meeting notes, and source code into ChatGPT to assist with debugging and documentation. The incidents were only identified after Samsung issued a survey to staff following OpenAI's disclosure of the Redis vulnerability — at which point engineers self-reported. Samsung immediately banned ChatGPT and other external generative AI tools company-wide. The data submitted had potentially been used in OpenAI's model training pipeline, and Samsung had no contractual basis to demand its deletion or verify its handling.

Samsung was not breached in the traditional sense. Its own employees made rational, productivity-maximising decisions — and in doing so inadvertently transferred some of the company's most sensitive intellectual property to a third-party AI training corpus, with no retrieval mechanism and no notice to legal or compliance.

The Four Fates of Data Submitted to AI Vendors

When an employee or system sends data to an external AI service, that data typically follows one of four pathways — and which pathway applies depends entirely on contract terms that most users never read:

1. Inference-only, no retention: The data is processed to generate a response and immediately discarded. This is the regime major enterprise API contracts (OpenAI API, Azure OpenAI, Anthropic API, Google Vertex AI) now offer by default — but it must be explicitly verified in contract terms and is typically not the default for free consumer tiers.

2. Retained for safety review: The data is stored for a defined period (often 30 days) for abuse detection and safety monitoring by the vendor. This is standard even in enterprise API agreements. The data is not used for training but is accessible to vendor trust-and-safety staff.

3. Retained and used for model improvement: The data is used to fine-tune or retrain future model versions. This was OpenAI's default for ChatGPT free and Plus users until April 2023, when opt-out was introduced. Many smaller AI vendors still default to this regime. Samsung's incident fell into this category.

4. Shared with subprocessors for operational purposes: The data is passed to the vendor's own subprocessors — cloud infrastructure providers, monitoring services, human labelling contractors — for operational purposes. These subprocessors may have their own retention and data rights terms.

Real Case: Italy's GDPR Action Against ChatGPT

In March 2023, Italy's data protection authority (Garante) temporarily banned ChatGPT in Italy on GDPR grounds, citing the lack of a legal basis for processing personal data of EU residents and the absence of age verification. OpenAI was given 20 days to respond. The ban was lifted in late April 2023 after OpenAI added privacy controls, opt-out mechanisms, and a transparency notice — but the episode demonstrated that cross-border AI data flows create regulatory exposure for enterprises even when the vendor, not the enterprise, controls the processing.

Cross-Border Data Transfer Risk in AI Pipelines

Foundation model inference infrastructure is concentrated in the United States (primarily AWS us-east-1, us-west-2, and Azure eastus regions). This means that even when an enterprise is domiciled in the EU, UK, or Australia, data submitted to US-based AI vendors is typically processed in the US, triggering cross-border transfer obligations under GDPR, the UK GDPR, or Australia's Privacy Act.

The standard legal mechanism is a Data Processing Agreement (DPA) incorporating Standard Contractual Clauses (SCCs for EU transfers). OpenAI, Google, Microsoft, and Anthropic all offer enterprise DPAs. However, in practice, the majority of enterprise deployments of AI tools are made by individual teams or functions that bypass central procurement — meaning no DPA exists, no transfer mechanism is in place, and the organisation may be in ongoing breach of data protection law for transfers it does not even know are occurring.

A 2023 survey by the law firm Fieldfisher found that 67% of European enterprises that had deployed AI tools at the departmental level had not completed the required Transfer Impact Assessments for US-based model providers. This is a systemic compliance gap, not an outlier problem.

Intellectual Property Contamination Risk

Beyond data protection law, AI data flows create a distinct intellectual property risk. When proprietary code, trade secrets, client data, or confidential strategy documents are submitted to an AI vendor whose terms permit use for training, that information may surface — transformed, but potentially recognisable — in responses to future users. This is not merely theoretical: GitHub Copilot has been documented reproducing verbatim code segments from private repositories where the repository owner's licence terms permitted training use.

The IP contamination risk runs in both directions. Code or content generated by AI tools may itself incorporate training data from copyrighted sources, potentially creating infringement exposure for the organisation that deploys the output. The ongoing litigation between The New York Times and OpenAI (filed December 2023) turns precisely on this mechanism.

Business Leader Takeaway

Before any AI tool touches business data, your organisation must determine: (1) which of the four data fate categories applies under the vendor's current contract terms, (2) whether a signed DPA with appropriate transfer mechanisms is in place, (3) whether the data type is permissible under those terms, and (4) whether IP generated by or submitted to the tool creates downstream ownership or liability risk. Answering these questions after deployment is dramatically more expensive than before.

Lesson 2 Quiz

3 questions — free, untracked, retake anytime.
What made the Samsung ChatGPT incident in March 2023 particularly significant from a supply chain risk perspective?
✓ Correct. The risk was not a hack — it was employees making rational productivity decisions that inadvertently transferred sensitive IP to a vendor whose terms permitted training use. Samsung had no contractual right to retrieve or delete the submitted data.
✗ The incident was not a hack. Employees voluntarily submitted proprietary chip designs and source code to ChatGPT for legitimate work purposes — but the terms at the time permitted OpenAI to use that data for training, and Samsung had no retrieval mechanism.
Under current enterprise API agreements from major AI providers, what is typically the default data handling regime for submitted content?
✓ Correct. Enterprise API contracts from OpenAI, Anthropic, Google, and Microsoft now typically default to inference-only processing with no retention for training — but this must be verified in the specific contract and is not the default for consumer tiers.
✗ Enterprise API contracts from major providers (OpenAI, Anthropic, Google, Microsoft) now default to inference-only, no-retention-for-training regimes — but only under signed enterprise agreements, not consumer or free-tier services.
The 2023 Fieldfisher survey found that approximately what percentage of European enterprises deploying departmental AI tools had NOT completed the required Transfer Impact Assessments for US-based AI providers?
✓ Correct. 67% — a majority — of surveyed European enterprises with departmental AI deployments had not completed Transfer Impact Assessments, representing systemic compliance exposure rather than isolated incidents.
✗ The Fieldfisher survey found 67% of European enterprises with departmental AI deployments had not completed required Transfer Impact Assessments, indicating a widespread systemic compliance gap.

Lab 2: AI Data Flow Risk Assessment

Classify your data flows, identify regulatory gaps, and draft DPA checklist items.

Assessing Where Your Data Goes in AI Pipelines

In this lab you will work through the practical challenge of understanding what data your organisation sends to AI vendors, how it is handled, and whether appropriate legal mechanisms are in place. You will practice categorising AI data flows and identifying compliance gaps before they become regulatory incidents.

Describe a specific AI tool or workflow at your organisation — what data it processes and which vendor it uses — and the assistant will help you identify which of the four data fate categories applies, what transfer obligations are triggered, and what due diligence steps are outstanding.

Try asking: "Our marketing team uses an AI writing tool that is connected to our CRM and processes customer email data. The tool is US-based. What data risks do we need to assess?" — or describe your own scenario.
AI Data Flow Risk Lab Lab 2
AI Risk for Business Leaders · Module 5 · Lesson 3

Model Poisoning, Trojan Models, and the Integrity of Your AI Supply Chain

The most dangerous supply chain attacks target the model itself — before it ever reaches your organisation.

In 2021, researchers from the University of Maryland and UC Santa Barbara published findings on what they termed "Hidden Killer" — a training-data poisoning attack capable of embedding targeted backdoors in NLP models that activated only on specific trigger phrases, while leaving the model's performance on standard benchmarks completely normal. The attack required the adversary to poison only 0.2% of the training dataset — roughly one in five hundred examples — to achieve near-perfect backdoor trigger reliability.

The paper's most commercially significant finding was that the poisoning was undetectable by standard model evaluation. An organisation that downloaded a poisoned open-source model from Hugging Face, evaluated it on standard test sets, and deployed it in production would have no indication the backdoor existed until an adversary sent the trigger phrase. At that point, the model would output whatever the attacker had specified — incorrect medical advice, fraudulent financial guidance, or manipulated code.

The Open-Source Model Risk Surface

The rapid proliferation of open-source foundation models — Meta's Llama series, Mistral, Falcon, Bloom, and hundreds of fine-tuned derivatives — has created a new and largely unmanaged risk surface for enterprise AI. As of 2024, Hugging Face's model hub hosts over 700,000 publicly available models, the vast majority of which have undergone no independent security evaluation.

Organisations that choose open-source models to reduce vendor dependency or cost often pull models from Hugging Face or similar repositories without verifying the provenance of the training data, the identity of the model uploader, or whether the model has been subjected to any form of red-teaming or security audit. This is the AI equivalent of downloading executable software from an anonymous forum and running it in your production environment.

The risk is not merely theoretical. In 2023, security researchers at Protect AI discovered that models uploaded to Hugging Face could execute arbitrary code during the deserialization of model files — a vulnerability in the pickle file format used by PyTorch. Malicious model files had been uploaded to the platform. Protect AI's scanning tool identified over 2,800 potentially malicious models on the platform before coordinated disclosure prompted Hugging Face to add security scanning. Organisations that had downloaded and deployed those models had remote code execution vulnerabilities in their AI infrastructure.

Real Case: SolarWinds Parallel in AI — The XZ Utils Incident

In April 2024, a sophisticated multi-year supply chain attack on the XZ Utils open-source compression library was discovered by a Microsoft engineer, Andres Freund, only by accident — he noticed slightly elevated SSH login times. The attacker had spent two years building credibility as an open-source contributor before inserting a backdoor. AI and ML frameworks depend heavily on the same open-source ecosystem of libraries. The XZ Utils case confirmed that nation-state actors are actively investing in long-horizon supply chain compromise of open-source software — the AI model ecosystem is not immune.

Prompt Injection via Third-Party AI Plugins and Tools

As enterprises deploy AI agents that can browse the web, read emails, execute code, and interact with external APIs, a new class of supply chain attack has emerged: indirect prompt injection. In this attack, malicious instructions are embedded in content that the AI system reads — a webpage, a document, an email — and those instructions hijack the AI's subsequent actions.

In 2023, security researchers demonstrated that an AI assistant with access to a user's email could be instructed, via text hidden in a received email, to forward all future emails to an attacker's address. The user sees nothing. The AI simply executes the instruction it found in the third-party content. This attack vector is entirely in the supply chain of content the AI system processes — not in the AI system itself.

The practical implication for organisations deploying AI agents with tool-use capabilities (Microsoft Copilot for M365, Google Workspace AI, Salesforce Einstein Agents) is that every external data source the agent reads becomes a potential attack surface. Trust boundaries must be architected into agent workflows before deployment, not patched in afterward.

Evaluating Model Integrity Before Deployment

Responsible AI procurement, whether of open-source models or fine-tuned vendor models, should include formal model integrity assessment. Current best-practice elements include:

Provenance verification: Can the vendor document the origin of training data, the training compute environment, and the chain of custody of model weights? NIST's AI Risk Management Framework (AI RMF 1.0, January 2023) lists provenance as a core supply chain risk management requirement.

Backdoor red-teaming: Has the model been subjected to adversarial testing specifically designed to elicit hidden behaviours under trigger conditions? This is distinct from standard performance benchmarking and requires specialist capability.

Dependency scanning: Are the software libraries used to serve the model (inference frameworks, serialization formats) free of known vulnerabilities? The Protect AI findings confirm this is not a theoretical concern for production deployments.

Business Leader Takeaway

The integrity of an AI model cannot be assumed from benchmark performance alone. Organisations deploying open-source or vendor-fine-tuned models must treat model provenance and integrity verification as procurement requirements equivalent to penetration testing in traditional software procurement — and must architect explicit trust boundaries into any AI agent that processes external content.

Lesson 3 Quiz

3 questions — free, untracked, retake anytime.
The "Hidden Killer" research demonstrated that a training-data poisoning attack could embed undetectable backdoors by corrupting what fraction of training data?
✓ Correct. The Hidden Killer research showed that poisoning only 0.2% of training data — roughly one in five hundred examples — was sufficient for near-perfect backdoor trigger reliability, while leaving standard benchmark performance unaffected.
✗ The Hidden Killer attack required poisoning only 0.2% of the training dataset — approximately one in five hundred examples — to achieve near-perfect backdoor activation, while benchmark performance remained normal and gave no indication of compromise.
What specific technical vulnerability did Protect AI discover in models on Hugging Face's platform in 2023?
✓ Correct. Protect AI found that the pickle serialization format used by PyTorch could execute arbitrary code when model files were loaded — and malicious models exploiting this had been uploaded to Hugging Face. Over 2,800 potentially malicious models were identified.
✗ Protect AI discovered that the pickle file format used by PyTorch to store model weights could execute arbitrary code during deserialization. Malicious models exploiting this vulnerability had been uploaded to Hugging Face, creating remote code execution risk for any organisation that downloaded and loaded them.
What is "indirect prompt injection" as demonstrated in 2023 security research?
✓ Correct. Indirect prompt injection embeds attack instructions in external content the AI processes — a webpage, email, or document — rather than in the direct user prompt. The AI then executes the hidden instruction, as demonstrated by researchers causing email-connected AI assistants to exfiltrate mailbox data.
✗ Indirect prompt injection places the attack instructions in content the AI reads from external sources — a webpage, email, or document — not in the direct user prompt. The AI then follows those hidden instructions, often without any visible indication to the user.

Lab 3: Model Integrity and Prompt Injection Risk

Practise assessing open-source model risk and designing agent trust boundaries.

Evaluating AI Model Integrity and Agentic Risk

In this lab you will practise identifying model integrity risks in open-source AI procurement and designing appropriate trust boundaries for AI agent deployments. You will work through real evaluation criteria and think through how indirect prompt injection attacks could affect specific agentic AI workflows at your organisation.

Describe an open-source model you are considering, or an AI agent workflow your organisation uses or is planning, and the assistant will guide you through the integrity assessment questions and trust boundary design considerations.

Try asking: "We're considering deploying a fine-tuned Llama 3 model that we downloaded from Hugging Face for customer support. What integrity checks should we run before putting it in production?" — or describe your own scenario.
Model Integrity Lab Lab 3
AI Risk for Business Leaders · Module 5 · Lesson 4

Governing AI Vendor Risk: Contracts, Audits, and Incident Response

Vendor risk management that was adequate for SaaS procurement is systematically insufficient for AI.

In 2023, UK insurer Aviva disclosed in its annual report that it had established a formal AI Vendor Risk Committee following an internal review that found 23 AI tools deployed across business units, only four of which had gone through standard vendor risk assessment. The remaining 19 had been procured by individual teams, often under "innovation" budgets that bypassed central IT and legal review. Two of the 19 had no signed data processing agreements. One was processing claims data through a US-based AI vendor with no GDPR transfer mechanism in place.

Aviva's public disclosure was notable because it was voluntary — the company chose to acknowledge the gap rather than wait for a regulatory finding. The remediation programme cost significantly more than prevention would have: renegotiating contracts retrospectively, conducting Transfer Impact Assessments under time pressure, and managing the reputational dimension of disclosing to regulators that claims data had been processed without adequate legal basis.

What AI Vendor Contracts Must Now Include

Standard SaaS vendor contracts were not designed for AI. They typically address data processing, security standards, and service levels — but do not address model-specific risks. An AI vendor contract should now include provisions covering:

Model version stability: Commitment to advance notice (minimum 90 days is emerging as standard) before model version deprecation or significant capability changes. Changes to model behaviour can invalidate downstream applications even without a security incident.

Training data usage rights: Explicit prohibition on using submitted data for model training or improvement, with audit rights to verify compliance. The prohibition should extend to all subprocessors.

Subprocessor disclosure and approval: Full list of all AI-specific subprocessors (not just generic IT subprocessors) and contractual right to object to additions. The list should include the specific foundation model provider(s) used.

Model incident notification: Obligation to notify the customer within a defined window (72 hours is the GDPR standard for personal data breaches) of any model behaviour incident, training data incident, or infrastructure compromise that could affect model outputs or data handling.

Right to audit AI governance: Right to receive third-party AI audit reports (analogous to SOC 2 for security) on at least an annual basis. The EU AI Act creates a formal audit requirement for high-risk AI systems; contracts with vendors deploying in the EU should reference this framework.

Regulatory Development: EU AI Act Supply Chain Obligations

The EU AI Act (entered into force August 2024) creates explicit supply chain obligations for AI. Providers of high-risk AI systems must maintain technical documentation including training data characteristics, system architecture, and validation results. Deployers of third-party high-risk AI systems must conduct conformity assessments of those systems before deployment. For the first time, the law treats AI supply chain governance as a legal requirement rather than a best practice — and penalties for non-compliance can reach 3% of global annual turnover.

AI-Specific Incident Response Planning

Traditional incident response plans address security breaches and system outages. AI supply chain incidents create a distinct category that most IR plans do not address: model behaviour incidents, where the AI system continues to function technically but is producing systematically incorrect, biased, or manipulated outputs due to a supply chain compromise or model change.

A model behaviour incident may be invisible to standard monitoring — the system is "up," API calls are completing, error rates are normal. Detection requires output monitoring: continuous statistical comparison of model outputs against baseline distributions, flagging of unexpected shifts in output characteristics. This is a capability most organisations deploying AI tools have not yet built.

In 2023, researchers documented that GPT-4's performance on several coding and mathematics benchmarks statistically declined between March and June 2023, as measured by Stanford and UC Berkeley researchers in a paper titled "How Is ChatGPT's Behavior Changing over Time?" The cause was not disclosed by OpenAI. The practical implication is that AI systems in production can silently degrade — and organisations without output monitoring will not detect the degradation until business impact surfaces through other channels.

Building a Shadow AI Registry

The Aviva case illustrates a near-universal problem: shadow AI — AI tools deployed without central visibility — is the rule, not the exception. A 2023 survey by Salesforce found that 55% of employees were using AI tools that had not been approved by their employer's IT function. This shadow deployment creates a risk register that the organisation cannot see, manage, or respond to.

Organisations that have successfully addressed this have typically combined three measures: (1) a mandatory AI tool registry requiring any team deploying an AI tool to submit a brief standardised declaration covering the tool, vendor, data types processed, and contract status; (2) network monitoring to detect traffic to known AI service endpoints, creating visibility into unapproved deployments; and (3) a fast-track approval process that takes no more than five business days for low-risk AI tools, removing the incentive to bypass procurement by making compliant procurement genuinely fast.

Business Leader Takeaway

Governing AI vendor risk requires extending your vendor risk management framework in three directions simultaneously: contracts that address model-specific risks, incident response plans that cover model behaviour incidents (not just security breaches), and registry processes that create visibility into shadow AI deployments before they become compliance events. The cost of remediation consistently exceeds the cost of prevention by a factor of three to ten — as Aviva's experience demonstrates.

Lesson 4 Quiz

3 questions — free, untracked, retake anytime.
What did the 2023 Stanford/UC Berkeley research paper "How Is ChatGPT's Behavior Changing over Time?" document?
✓ Correct. The research found measurable statistical decline in GPT-4's coding and mathematics performance between March and June 2023 — a silent degradation that organisations without output monitoring would not have detected until business impact surfaced.
✗ The research documented a measurable statistical decline in GPT-4's performance on coding and mathematics benchmarks between March and June 2023, without any disclosure or explanation from OpenAI. This illustrates how AI systems can silently degrade in ways invisible to standard monitoring.
Under the EU AI Act (entered into force August 2024), what is the maximum penalty for supply chain compliance failures by providers of high-risk AI systems?
✓ Correct. The EU AI Act sets penalties up to 3% of global annual turnover for failures to comply with obligations including supply chain documentation, conformity assessment, and technical documentation requirements for high-risk AI systems.
✗ The EU AI Act sets penalties of up to 3% of global annual turnover for non-compliance with its supply chain and conformity assessment obligations — making AI supply chain governance a legally mandated requirement with significant financial consequences in the EU.
What three-measure combination has been most effective at addressing shadow AI deployments in enterprise organisations?
✓ Correct. The combination of mandatory registry (visibility), network monitoring (detection), and fast-track approval (removing the incentive to bypass) addresses shadow AI at the root cause — the friction of compliant procurement — rather than only at the enforcement level.
✗ The most effective combination pairs mandatory registry and network monitoring (to create and detect visibility) with a genuinely fast approval process — removing the incentive to bypass procurement by making compliant procurement faster than going around it.

Lab 4: AI Vendor Governance and Incident Response

Draft contract provisions, build a shadow AI registry, and design an AI incident response playbook.

Designing Your AI Vendor Governance Framework

In this lab you will practise the governance design skills required to manage AI vendor risk at an organisational level: drafting specific contract provisions, designing a practical shadow AI registry process, and building AI-specific incident response procedures that address model behaviour incidents — not just security breaches.

Describe your organisation's current AI vendor governance situation — what you have in place and what gaps you are aware of — and the assistant will help you prioritise and draft specific governance artefacts appropriate to your context.

Try asking: "We're a 5,000-person financial services firm. We have a SOC 2-based vendor review process but no specific AI provisions. We've found about 12 AI tools deployed without central visibility. Where do I start?" — or describe your own governance challenge.
AI Governance Lab Lab 4

Module 5 Test

15 questions. Score 80% or above to pass.
1. The March 2023 ChatGPT data exposure that revealed user chat titles and payment information was caused by what?
✓ Correct. The incident originated in a third-party Redis library — a textbook supply chain vulnerability propagating upstream to end users.
✗ The vulnerability was in a third-party Redis caching library — a supply chain component, not OpenAI's own code.
2. "Provider concentration risk" in AI refers to which of the following?
✓ Correct. Provider concentration risk means that disruption, governance failure, or policy change at a single foundation model provider cascades as a business continuity event across all dependent AI applications.
✗ Provider concentration risk refers to the business continuity exposure created by over-reliance on a single AI infrastructure provider — illustrated by the November 2023 OpenAI governance crisis.
3. Which November 2023 event prompted banks including Goldman Sachs and JPMorgan to formally document AI provider concentration risk in internal risk registers?
✓ Correct. The five-day OpenAI governance crisis revealed how dependent enterprises had become on a single provider's continuity — triggering formal concentration risk documentation at major financial institutions.
✗ The November 2023 OpenAI governance crisis — the board's firing and reinstatement of Sam Altman — was the trigger, exposing how governance instability at a vendor creates downstream enterprise business continuity risk.
4. When proprietary Samsung chip schematics and source code were submitted to ChatGPT in March 2023, what was the primary supply chain risk dimension?
✓ Correct. The risk was that employees voluntarily submitted sensitive IP to a vendor whose terms permitted training use — and Samsung had no right under those terms to retrieve or delete the data.
✗ The risk was that employees, making rational productivity decisions, submitted IP to a service whose terms permitted training data use — creating irreversible IP transfer with no contractual remedy.
5. Italy's data protection authority (Garante) temporarily banned ChatGPT in March 2023 on what primary legal basis?
✓ Correct. The Garante cited GDPR violations including lack of legal basis for processing EU residents' personal data and no age verification — demonstrating that cross-border AI data flows create regulatory exposure even when the enterprise is not the one controlling the processing.
✗ The Garante cited lack of legal basis for processing EU residents' personal data and absence of age verification as the primary GDPR violations — illustrating that cross-border AI data flows create regulatory risk for deploying organisations regardless of who controls the processing.
6. What are "Standard Contractual Clauses" (SCCs) in the context of AI vendor data flows?
✓ Correct. SCCs are EU-approved contract clauses that provide a legal mechanism for transferring personal data to countries outside the EEA that lack adequacy decisions — essential for transfers to US-based AI infrastructure providers.
✗ Standard Contractual Clauses are EU-approved legal mechanisms for transferring personal data to non-adequate countries like the US. They must be incorporated into Data Processing Agreements with AI vendors processing EU residents' data.
7. The "Hidden Killer" backdoor poisoning research found that standard model evaluation benchmarks were effective at detecting the embedded backdoors. True or false?
✓ Correct. The paper's most significant finding was that the backdoors were completely undetectable by standard benchmark evaluation — poisoned models performed normally on test sets, with no indicator of compromise until the trigger phrase was used.
✗ False. The critical finding was that standard benchmark evaluation failed completely to detect the backdoors. The model performed normally on all standard test sets, making detection impossible without specialist adversarial red-teaming targeting the specific trigger mechanism.
8. Protect AI's security research on Hugging Face in 2023 identified what specific threat to organisations downloading models from the platform?
✓ Correct. Protect AI found that PyTorch's pickle serialization format could be exploited in malicious model files to execute arbitrary code — and over 2,800 potentially malicious models had already been uploaded to Hugging Face before coordinated disclosure.
✗ Protect AI discovered that malicious model files could exploit PyTorch's pickle deserialization format to execute arbitrary code when models were loaded — a supply chain attack that over 2,800 Hugging Face models were potentially exploiting.
9. In an indirect prompt injection attack demonstrated by security researchers in 2023, what action did the AI assistant perform after reading a maliciously crafted email?
✓ Correct. The demonstrated attack caused the AI email assistant to silently forward all future mailbox contents to the attacker — triggered entirely by hidden instructions in received email content, with no user awareness.
✗ Researchers demonstrated that hidden instructions in a received email could cause an AI email assistant to silently forward all future mailbox contents to an attacker — without any visible indication to the user that anything had occurred.
10. What is the minimum advance notice period that is emerging as standard in AI vendor contracts for model version deprecation or significant capability changes?
✓ Correct. 90 days is emerging as the standard minimum notice period for model deprecation — consistent with the notice OpenAI gave for GPT-3.5/Codex deprecation in 2023. This gives organisations time to migrate dependent applications.
✗ 90 days is emerging as the standard minimum. OpenAI's 2023 deprecation of the original GPT-3.5 and Codex endpoints established this precedent, though it was considered the minimum rather than adequate notice for organisations with complex dependent applications.
11. The EU AI Act (entered into force August 2024) treats AI supply chain governance as what?
✓ Correct. The EU AI Act creates binding legal supply chain obligations for high-risk AI systems, including technical documentation, conformity assessment, and subprocessor transparency — backed by penalties up to 3% of global annual turnover.
✗ The EU AI Act makes AI supply chain governance a binding legal requirement for high-risk AI systems, with penalties reaching 3% of global annual turnover — transforming what was previously a best practice into an enforceable legal obligation.
12. A 2023 Salesforce survey found that what percentage of employees were using AI tools not approved by their employer's IT function?
✓ Correct. The Salesforce survey found 55% of employees using unapproved AI tools — confirming that shadow AI is a majority-prevalence phenomenon requiring systematic governance rather than individual enforcement.
✗ Salesforce's 2023 survey found 55% of employees were using AI tools not approved by IT — indicating shadow AI is not an edge case but a majority-prevalence phenomenon requiring systematic governance response.
13. What is a "model behaviour incident" as distinct from a traditional IT security incident?
✓ Correct. A model behaviour incident is invisible to standard IT monitoring — the system is "up" and API calls complete — but outputs are systematically wrong, degraded, or manipulated. Detection requires output monitoring, not infrastructure monitoring.
✗ A model behaviour incident occurs when the AI system is technically operational but producing systematically incorrect or degraded outputs — invisible to standard IT monitoring and detectable only through output statistical monitoring against baseline distributions.
14. What specific NIST document (published January 2023) lists AI supply chain provenance verification as a core risk management requirement?
✓ Correct. NIST's AI RMF 1.0, published January 2023, is the primary US federal framework for AI risk management and specifically lists supply chain provenance verification — including training data origin and model weight chain of custody — as a core requirement.
✗ The NIST AI Risk Management Framework (AI RMF 1.0), published in January 2023, is the correct document. It specifically addresses AI supply chain risk including training data provenance and model integrity verification as core risk management requirements.
15. The XZ Utils supply chain attack discovered in April 2024 is relevant to AI supply chain risk because it demonstrated what?
✓ Correct. The XZ Utils case — a two-year patient supply chain attack by a nation-state actor on a foundational open-source library — confirmed that the open-source ecosystem on which AI frameworks depend is an active target for sophisticated long-horizon supply chain attacks.
✗ XZ Utils demonstrated that nation-state actors are conducting multi-year, patient supply chain attacks on the open-source software ecosystem — the same ecosystem of libraries and dependencies that AI and ML frameworks rely upon, making AI infrastructure directly exposed to this threat class.