AI Security Landscape

As organisations increasingly adopt artificial intelligence, new security challenges emerge that differ significantly from traditional cybersecurity threats. AI systems introduce unique attack surfaces and require specialised governance frameworks.

Key AI Security Threats

Prompt Injection

Attackers craft malicious inputs to manipulate the behaviour of large language models, potentially causing the system to reveal sensitive information, generate harmful content, or perform unauthorised actions.

Data Poisoning

Introducing malicious or misleading data into training datasets to corrupt model behaviour. This can lead to biased outputs, reduced accuracy, or intentional backdoors in the model.

Model Theft

Unauthorised copying or reverse-engineering of AI models through API queries or other means. This represents both an intellectual property concern and a security risk if the model contains sensitive training data.

Adversarial Examples

Carefully crafted inputs designed to cause AI models to make specific errors. In security contexts, this could mean evading malware detection or misclassifying threats.

AI Governance Framework

Hong Kong’s AI Taskforce has proposed a governance framework built on five pillars:

  1. Accountability: Clear responsibility for AI system outcomes
  2. Transparency: Disclosure of AI usage and decision-making processes
  3. Fairness: Prevention of bias and discrimination in AI outputs
  4. Reliability: Ensuring AI systems perform as intended across scenarios
  5. Security: Protecting AI systems from manipulation and abuse

Practical Recommendations

  • Implement input validation and sanitisation for all AI system inputs
  • Maintain comprehensive logs of AI system interactions
  • Conduct regular security assessments of AI models and pipelines
  • Establish human oversight for high-stakes AI decisions
  • Develop incident response procedures specific to AI security events

Hong Kong operator checklist

  1. Confirm whether the systems, vendors, or practices described apply to your estate.
  2. Assign an owner and a review date — do not leave findings as unread newsletter content.
  3. Capture evidence (configs, tickets, screenshots) if you later enter a Trust Review.
  4. Brief leadership with a dated one-page note when residual risk remains high.

What “good” looks like

  • Controls are operated, not only documented
  • Privileged access uses phishing-resistant MFA where feasible
  • Detection and response paths are exercised at least annually
  • Third-party dependencies have an owner and an exit plan

Sources and further reading

Editorial note

This page is published by the Hong Kong Information Security Group (HKISG) for educational and early-warning purposes. It is not a substitute for legal advice, formal audit opinions, or national CERT coordination.

Frequently asked questions

Who should read this?

Security, IT, and risk owners in Hong Kong organisations who need practical context rather than marketing claims.

Does this change any public HKISG rating?

No. TrustScores are produced only through the published Trust Review / Awards process. Reading this page does not alter scores.

How often is this content reviewed?

HKISG dates publications and retains corrections under our editorial standards. Check the updated field in the page header when present.