Course Objective & Scope
AI systems do not only fail by accident; they can be manipulated by malicious actors to leak confidential data, bypass business rules or generate harmful content. This course bridges AI ethics (unintended harm, manipulation, loss of human autonomy) and AI safety and security for business, compliance and risk leaders.
It covers the OWASP Top 10 for LLM Applications, direct and indirect prompt injection, jailbreaks, data poisoning, deepfake fraud and excessive agent agency through concrete business scenarios. Participants run a non-technical threat-modelling exercise on an organisational use case and specify guardrail and red-teaming requirements.
Who Should Attend?
- Information security (CISO), cyber governance and enterprise risk leaders
- Business owners launching customer-facing chatbots, copilots or AI agents
- Legal, compliance, internal control and fraud prevention teams
- Product and project managers defining safety requirements with engineering teams
Key Takeaways
- Distinguish and connect ethical harms, AI safety failures and adversarial security threats
- Interpret the OWASP Top 10 for LLM Applications in terms of business impact
- Spot direct and indirect prompt injection and excessive agent agency risks in workflows
- Commission and govern pre-launch red-teaming and adversarial testing exercises
- Specify input/output guardrails, least-privilege agent permissions and content filters
Curriculum
Unpacking AI Ethics, AI Safety and AI Security
- Unintended algorithmic harm versus systemic safety versus adversarial security attacks
- Manipulative design, cognitive overreliance and ethical red lines
- Deepfake voice/video impersonation, executive fraud and social engineering
OWASP Top 10 for LLM Applications for Business Leaders
- Direct and indirect prompt injection via untrusted emails, web pages or documents
- Sensitive information disclosure, insecure output handling and data poisoning
- Excessive agency: when tool-calling agents are tricked into unauthorised actions
Jailbreaks, Toxic Outputs and Brand Reputation Risks
- Preventing customer-facing bots from making binding promises or disparaging the brand
- Blocking off-topic discussions, competitor endorsements and legal/political traps
Layered Defence: Guardrails and Least-Privilege Architecture
- Input screening, system prompt hardening, output validation and PII redaction
- Enforcing read-only scopes, transaction limits and human approval for agents
Governing Red Teaming and Security Sign-Off
- Planning cross-functional red-teaming exercises before production launch
- Triaging adversarial findings and setting security release criteria
Other Courses in This Category
-
Build an enterprise AI governance framework that scales innovation safely: establish an AI system inventory, run risk-proportionate stage-gates across the lifecycle, and report risk and value to the board.
- Live Course
- 5 modules
-
Draft clear, enforceable corporate AI policies covering employee Generative AI Acceptable Use, data classification rules for prompts, AI procurement and development standards, and rollout communication.
- Live Course
- 5 modules
-
Close accountability gaps across AI systems by designing a Three Lines of Defence model and a clear RACI matrix linking business owners, AI product managers, model stewards, compliance, security and internal audit.
- Live Course
- 5 modules
-
Classify enterprise AI systems into risk tiers based on intended purpose, autonomy, affected stakeholders, data sensitivity and EU AI Act criteria, and map mandatory controls to each tier.
- Live Course
- 5 modules
-
Navigate the EU AI Act's extraterritorial impact, provider and deployer obligations, compliance milestones and penalties, and design a role-based corporate AI Literacy programme that satisfies Article 4.
- Live Course
- 5 modules
-
Manage KVKK and GDPR compliance in AI workflows, cross-border LLM API transfers, PII masking in prompts and RAG, copyright risks in training data, IP ownership of AI outputs, and protective contract clauses.
- Live Course
- 5 modules
Get in touch
