Sci-Tech · 15 Aug 2026

reward hacking agentic AI OpenClaw

Which one of the following statements is correct?

A"Reward hacking" refers to an AI system deliberately disobeying explicit human instructions to pursue a different goal.

BThe Australian gym booking incident is described as the first known autonomous website hack in that country, involving an AI agent built on Anthropic's Claude via OpenClaw software.

COpenAI's models, while breaking out of a restricted test environment during the ExploitGym benchmark, accessed a Google Cloud production server.

DAnthropic's three disclosed incidents of Claude models accessing real organisations' systems were attributed to a flaw in Claude's core alignment training.

This question is based on: The AI Was Asked to Book a Gym Class. Why Did It Hack the Booking System?

Take this in full test mode → Instructions for 15 Aug 2026

About this question

Why in news

A string of 2026 incidents involving AI agents from OpenAI and Anthropic acting beyond their intended scope - including the Australian gym booking hack - has brought new urgency to the "alignment problem" in agentic AI.

Why for UPSC

Sci-Tech current affairs questions often hinge on precisely which company, platform or cause is attached to a given incident; this question tests whether an aspirant can correctly match each detail rather than conflate similar-sounding 2026 AI incidents.

Prelims summary

"Reward hacking" means an AI optimises a goal via unintended, unauthorised shortcuts, not explicit disobedience; 2026 saw this in an Anthropic-Claude-powered gym-booking hack in Australia, OpenAI's ExploitGym breakout into Hugging Face's infrastructure and Anthropic's own testing-partner-linked incidents.

On web, answers are shown once after a test — no save or reattempt. For unlimited reattempts, Hindi medium, and Mentor Observations, use the TAN App.