Drewtech Series Chapter 20 - Will no one rid me of this turbulent priest? – AI agents and the boundaries of autonomy
18 Nov 2026
AI agents are given goals, not methods. When an agent pursues its goal by means that the instructing party never contemplated, who answers for the result? This article considers how Singapore law may attribute an agent’s unexpected conduct, why existing legal analogies fit imperfectly, and how organisations can allocate the risk by contract in the meantime.
Please click here to read more about the update.
If you missed any of the chapters in our DrewTech series, you can read them below:
- Chapter 1: The Importance of an Exit Strategy in Tech Contracts
- Chapter 2: Employees, technology and a legal hangover - bring your own problems?
- Chapter 3: I host, you post, I get sued?
- Chapter 4: Diabolus ex machina - Artificial (un)Intelligence and liability
- Chapter 5: Bringing Hygiene Online - The MAS Notice on Cyber Hygiene
- Chapter 6: Signing without signing – contactless contracts
- Chapter 7: My Kingdom for a Horse – When your Systems are Held to Ransom
- Chapter 8: New risks in new skins - Updates to the Guidelines on Risk Management Practices – Technology Risk
- Chapter 9: Of blockchains and stumbling blocks
- Chapter 10: Service by airdrop - no parachutes required
- Chapter 11: Large language models and larger legal minefields
- Chapter 12: Beset on all sides – liability for data breaches
- Chapter 13: Pitfalls of user-generated content
- Chapter 14: Red queen races – vulnerability disclosure programs
- Chapter 15: Looking at the man in the middle (in a cyber breach) – allocation of risk
- Chapter 16: Speak, friend, and enter - access controls and authorised users
- Chapter 17: I love the smell of AI in the morning – a review of an AI-generated letter of demand
- Chapter 18: Schrödinger’s Safe – Quantum Computing and the Law
- Chapter 19: When the lights go off – a review of key legal considerations for high-availability systems