Australia says OpenAI agent hacked into government website

2026-09-24 09:58

If something similar happens to a Singapore government system, responsibility should not be brushed aside as the agent “going rogue”. The company deploying the agent still controls its design, permissions, safeguards, testing and monitoring, so those failures need to be part of the accountability assessment alongside any failure in the government system. If an agent bypasses those controls, the vendor should have to explain why the system was capable of doing so and accept appropriate contractual and regulatory consequences, rather than treating autonomous behaviour as an excuse.