Anthropic settlement and infrastructure risk dominate AI roundup

A federal judge in San Francisco granted final approval to Anthropic’s $1.5B settlement with authors who accused the company of using pirated books to train Claude, marking the first major AI-training copyright case to resolve. The deal pays roughly $3,100 per work across more than 480,000 titles, with $122M set aside for plaintiffs’ attorney fees.

AI infrastructure remained a central theme. A Nikkei-cited study estimated Alphabet, Microsoft, Amazon, Meta and Oracle now carry $1.65T in off-balance-sheet debt tied to data-center leases and GPU supply contracts, up roughly 8x in about four years. Microsoft also became the first publicly named customer for AMD’s Helios rack-scale AI platform, deploying the 72-GPU MI455X system on Azure, with shipments beginning in H2 2026.

Security and policy developments widened the agenda. Pillar Security disclosed sandbox bypasses in Cursor, OpenAI Codex, Google Gemini CLI, and Google Antigravity, while the European Commission published guidance for AI Act transparency obligations that become enforceable August 2. Taiwan also indicted a former TSMC deputy manager in its first chip-secrets case tied to boosting Chinese semiconductor capabilities.