AI Trends — 3 August 2026

OpenAI's Astra Cracks Ten Open Math Problems

An internal version of OpenAI's next major model, Astra, reportedly solved ten previously unsolved problems in mathematics and theoretical computer science, with formal Lean proofs published on GitHub to verify the results — all for roughly $2,000 in compute. The results reportedly include a construction establishing the existence of non-sofic groups, a long-standing open question in group theory, and new upper bounds on sphere-packing density approaching the Cohn–Elkies threshold. If the claims hold up under scrutiny, it's a striking signal of how cheaply frontier reasoning capability is starting to compound.

EU Begins Enforcing AI Act Transparency Rules

Starting August 2, the European Commission began enforcing a new tranche of the EU AI Act, including transparency requirements for high-risk systems and generative AI. Chatbots and other interactive AI systems must now disclose that users are talking to AI rather than a human, and AI-generated or altered content — including deepfakes — must be labeled. It's the first continent-wide rule of its kind and a template regulators elsewhere are likely to study closely.

Buyer Attention Shifts From Demos to Working Tools

Industry coverage this week points to a broader shift in enterprise buying: less interest in flashy chat demos, more focus on tools that actually complete tasks. Google is pushing Gemini deeper into coding, search, and task-based workflows, while Meta and robotics vendors are seeing renewed attention as companies look for measurable productivity gains rather than novelty.

Security Researchers Flag DeepSeek-Powered Attack Campaign

Palo Alto Networks' Unit 42 detailed a Zhuhai-based threat actor, tracked as "knaithe," who wired DeepSeek into the open-source Hermes Agent framework and directed it via Telegram to enumerate targets, source public exploits, and attack more than 460 internet-facing systems. The campaign is one of the clearer public examples yet of an open-weight model being operationalized as an autonomous offensive tool, and is likely to sharpen scrutiny of agentic AI misuse.

Also Worth Noting

Agentic AI continues to dominate enterprise roadmaps, with analysts pointing to a sharp rise in task-specific agents shipping embedded in business applications this year, up from a small fraction just twelve months ago. Domain-specific, smaller models are also gaining ground as companies weigh efficiency and accuracy against the cost of running general-purpose frontier systems.

This newsletter was compiled from public reporting on AI developments over the past 48 hours.