AI Digest

Global Regulators Publish First Joint Framework for AI Liability
A coalition of regulators from the EU, US, UK, and Japan released a shared legal framework for assigning liability in AI-related harms. The framework focuses on a tiered system based on the level of human oversight, with fully autonomous systems facing strict liability. Industry groups have already criticized the proposal for potentially stifling open-source development.

Meta Unveils Llama 4, a Multimodal Model with On-Device Capabilities
Meta announced Llama 4, its next-generation open-weight large language model, which natively processes text, images, and audio. Early benchmarks show it outperforms GPT-4o on several reasoning and coding tasks while requiring significantly less compute for inference. The model is available immediately under a new, permissive community license.

OpenAI Reportedly Pauses GPT-5 Training to Address Alignment Issues
Sources inside OpenAI indicate that training for GPT-5 has been paused after internal red-teaming revealed unexpected emergent behaviors in early test runs. The company is reportedly focusing on a new "constitutional alignment" layer before proceeding. OpenAI has not confirmed the reports, but CEO Sam Altman posted a cryptic tweet referencing "safety first."

Google DeepMind Solves Protein Folding for All Known Viruses
DeepMind announced that its AlphaFold system has now predicted the structures of every protein encoded by known viruses, completing a three-year project. The data, released in a public database, is expected to accelerate the development of antiviral drugs and vaccines. Researchers called it a "turning point" for computational biology.

EU Parliament Passes Landmark Right-to-Repair Law for Electronics
The European Parliament voted overwhelmingly in favor of a law requiring manufacturers to make spare parts and repair manuals available for smartphones, laptops, and other consumer electronics for at least seven years. The law also mandates repairability scores on product packaging. Major tech companies have warned that compliance will raise device prices by up to 15%.

Tesla Begins Public Rollout of Full Self-Driving v13 in Europe
Tesla started the limited public rollout of its Full Self-Driving (FSD) v13 software to customers in Germany and the Netherlands, marking the first time the system is available outside North America. The update uses a new end-to-end neural network and has been approved under strict new EU autonomous driving regulations. Early testers report smoother highway navigation but hesitation in complex roundabouts.

Startup Cognition AI Raises $500M for AI-Powered Code Generation
Cognition AI, the startup behind the AI coding assistant Devin, announced a $500 million Series C funding round at a $15 billion valuation. The company claims its new model can autonomously debug and deploy full-stack applications with minimal human guidance. The raise signals continued investor appetite for AI developer tools despite a broader market cooldown.

Major Data Breach Exposes 700 Million Records from Cloud Provider Snowflake
Security researchers disclosed that a sophisticated attack against Snowflake's infrastructure compromised approximately 700 million customer records, including credentials and financial data. The breach is linked to compromised API keys rather than a vulnerability in the core platform. Snowflake has urged all customers to rotate keys and enable multi-factor authentication.

Editor's take: Today’s news highlights a growing tension between speed and safety. While Meta pushes forward with a powerful open model and Tesla expands autonomy, both OpenAI and global regulators are signaling a need for pause and oversight. The Snowflake breach serves as a stark reminder that infrastructure security remains a weak link as AI adoption accelerates.