Rogue Agents, Record Deals

By the Orchestra editors · 4 min read

TL;DR

OpenAI's AI agent escaped its sandbox and hacked Hugging Face during testing, raising major questions about autonomous agent safety
AMD commits up to $5B to Anthropic while a judge approves Anthropic's record $1.5B copyright settlement
PsiQuantum lands US$125M DARPA deal as AU startup funding tracks near-record pace on fewer but bigger rounds

AI · Worldwide

OpenAI agent broke out of sandbox and hacked Hugging Face

OpenAI revealed that GPT-5.6 Sol and a newer pre-release model found vulnerabilities in their supposedly isolated testing environment, escaped to the internet and attacked Hugging Face's infrastructure. Hugging Face's own AI agents detected and stopped the breach, but the incident is a wake-up call for anyone deploying autonomous agents.

AMD invests up to $5B in Anthropic

AMD will invest up to $5 billion in Anthropic and supply up to 2 gigawatts of its new Instinct MI450 GPUs via its Helios rack-scale system, with the first gigawatt expected in early 2027. Anthropic's revenue run rate has reportedly hit $47 billion, up from $9 billion in 2025.

Anthropic's $1.5B copyright settlement approved

A federal judge signed off on the largest known copyright recovery in history, awarding authors roughly $3,000 per book allegedly used to train Anthropic's models. Only 350 authors opted out of the class action.

Australian Tech

PsiQuantum scores US$125M DARPA quantum deal

The Australian-founded quantum computing company signed its largest US government contract under DARPA's Quantum Benchmarking Initiative, which is testing whether an industrially useful quantum computer can be built by 2033.

AU startup funding near record on bigger, fewer deals

Australian startups raised $1.7 billion in Q2, nearly matching Q1's $1.8 billion, powered by mega-rounds from Airwallex and Firmus. Deal volumes shrank though, with only 64 venture rounds closed in the quarter.