1. OpenAI's Astra Solves Ten Long-Standing Math Problems
OpenAI revealed that an internal version of its next major model, Astra, generated verified solutions to ten previously unsolved problems in mathematics and theoretical computer science, each open for a decade or more — including an explicit construction of a non-sofic group and a disproof of Connes's rigidity conjecture. Total compute cost: roughly $2,000. OpenAI published a 249-page manuscript plus Lean 4 proof certificates on GitHub with a zero "sorry" count, meaning every step is machine-verified.
2. OpenAI Pauses Astra Work Over Critical Cybersecurity Risk
Days after showcasing Astra's math results, OpenAI disclosed the model may have crossed the "Critical" threshold for cyber capability under its Preparedness Framework and paused internal work that doesn't yet meet stronger security controls. It's reportedly the first model to trigger that threshold, underscoring how fast capability gains are outpacing safety tooling.
3. Demis Hassabis Steps Down as Google DeepMind CEO
Demis Hassabis is stepping down as CEO of Google DeepMind to focus on AGI, moving into a Chair role at DeepMind and Chief Scientist at Alphabet while continuing to lead Isomorphic Labs. CTO Koray Kavukcuoglu takes over as SVP running DeepMind day-to-day. The announcement coincided with chief scientist Jeff Dean's departure to launch a new public-benefit AI corporation, and Alphabet shares dipped 5% on the news.
4. EU AI Act's High-Risk and Transparency Rules Take Effect
The EU AI Act moved from statute to daily enforcement this month: companies operating in Europe must now disclose when users are interacting with an AI system rather than a human, and providers of generative models must mark AI-generated output in a machine-readable way. It's the first major bloc-wide transparency mandate of its kind to bite in practice.
5. Mistral Releases Shieldstral, a 3B Policy-Adaptive Safety Classifier
Mistral AI open-sourced Shieldstral, a 3-billion-parameter multimodal safety classifier that judges text and images against plain-language moderation policies at inference time instead of a fixed harm taxonomy. Built on Ministral-3-3B with a Pixtral vision encoder, it runs on a single 16GB GPU, covers 12 languages, and Mistral says it matches guard models up to seven times its size on text safety.
6. Palantir Q2 2026 Revenue Up 93%, Guidance Raised Again
Palantir reported Q2 2026 revenue of $1.94 billion, up 92.8% year-over-year, with U.S. commercial revenue soaring 149%. The company raised full-year guidance to $8.15–8.16 billion (82% growth) and posted a Rule-of-40 score of 155%, extending its run as the bellwether stock for enterprise AI deployment.
7. NVIDIA Open-Sources NOOA, a Single-Class Agent Framework
NVIDIA released NOOA (NVIDIA Object-Oriented Agents), a model-agnostic Python framework that collapses agent development into one class — methods are actions, docstrings are prompts, type annotations are enforced contracts. NVIDIA reports 82.2% on SWE-bench Verified and 86.8% on CyberGym L1 at roughly half the token cost of comparable open harnesses. Shipped Apache 2.0, alpha preview, pip-installable.
8. Chip Capacity Becomes the Real AI Bottleneck: TSMC and Tesla/SpaceX Expand
TSMC added another $100 billion to its Arizona buildout, pushing total U.S. investment to $265 billion and raising 2026 capex guidance to $60–64 billion. Separately, Tesla and SpaceX unveiled a $16.8 billion joint chip plant near Houston, dubbed "Terafab." Analysts increasingly frame late-2026 AI progress as gated by semiconductor and power infrastructure rather than model research.
// KEY TAKEAWAYS
The frontier lab story this week split in two directions: OpenAI showed Astra doing genuine novel research, then immediately hit its own safety brakes on the same model — a rare public instance of a lab pausing itself. Meanwhile leadership churn (Hassabis stepping back at DeepMind) and hard infrastructure spend (TSMC, Tesla/SpaceX) suggest the industry's bottleneck is shifting from clever algorithms to chips, power, and governance — with the EU's transparency rules now testing whether regulation can keep pace with either.