AI Digest

Global AI Safety Treaty Ratified by 42 Nations
A landmark international agreement on AI safety was formally ratified today by 42 countries, including the US, China, and the EU. The treaty mandates mandatory risk assessments for advanced models and establishes a shared incident reporting system for catastrophic failures. Signatories will begin domestic implementation over the next 12 months.

DeepMind Unveils AlphaFold 3, Predicting All Molecular Interactions
DeepMind released AlphaFold 3, a significant leap beyond protein folding that can now predict the structure of nearly all molecular interactions, including DNA, RNA, and small molecule ligands. The model is expected to dramatically accelerate drug discovery and materials science. The company is making the core code available to academic researchers under a new open license.

OpenAI Launches GPT-5 with Enhanced Reasoning and Tool Use
OpenAI formally launched GPT-5, its next-generation large language model, featuring significantly improved reasoning chains, better long-context memory, and native integration with external APIs for real-time actions. Early benchmarks show a 40% reduction in factual errors compared to GPT-4 Turbo. The model is rolling out to all ChatGPT Plus subscribers today.

EU Fines Meta €1.2 Billion Over Data Transfer Violations
The European Union levied a record €1.2 billion fine against Meta for illegally transferring user data from Europe to the United States in violation of GDPR. Regulators cited the company’s failure to implement adequate safeguards for EU citizens' data used in training AI models. Meta stated it will appeal, calling the fine "unjustified and excessive."

Apple and Google Announce Joint Standard for On-Device AI Privacy
Apple and Google jointly announced a new technical standard, "PrivateML," designed to allow AI models to run sensitive tasks on-device without sending raw data to the cloud. The standard includes a common framework for differential privacy and encrypted inference. Both companies plan to adopt PrivateML in their next mobile operating system updates this fall.

Anthropic Releases Claude 3.5 Opus, Claiming Top Coding Performance
Anthropic released Claude 3.5 Opus, a new flagship model that it claims surpasses GPT-5 on several coding and mathematics benchmarks. The model introduces a "Self-Correction" mode that allows it to review and refine its own output before finalizing a response. It is available now via the Anthropic API and the Claude Pro subscription tier.

US Department of Justice Sues Nvidia Over AI Chip Market Monopoly
The US Department of Justice filed an antitrust lawsuit against Nvidia, alleging the company uses illegal bundling practices and exclusive contracts to maintain a monopoly in the AI accelerator market. The suit seeks to break Nvidia’s dominant position by requiring it to separate hardware sales from its CUDA software ecosystem. Nvidia called the lawsuit "baseless" and vowed to fight it.

Tesla Begins Full Rollout of FSD v13.0 to All US Customers
Tesla began a wide-scale rollout of its Full Self-Driving (Supervised) v13.0 software to all US customers who have purchased the option. The update is the first to rely entirely on a "vision-only" neural network, removing all radar and ultrasonic sensor dependencies. Early driver reports indicate improved highway lane changes and smoother urban navigation, though a "hands-on-wheel" requirement remains.

Editor's take: Today’s news is dominated by a single theme: the end of the hands-off era. Between the new global safety treaty, the EU’s record fine, and the DoJ’s antitrust suit against Nvidia, it is clear that governments and regulators are moving aggressively to impose structure—on data, on market power, and on safety. The simultaneous release of GPT-5, Claude 3.5 Opus, and AlphaFold 3 signals that the technology is accelerating faster than ever, making these legal and policy actions both urgent and inevitable.