21 Sep 2025

AI News Digest

🤖 AI-curated 7 stories

Today's Summary

Meta’s getting serious with AR, rolling out their Ray-Ban Display smart glasses that come with a wristband for gesture controls. This move is a clear sign they’re not just dreaming about AI living in your phone but in your everyday eyewear, which could change how we interact with digital info. Meanwhile, Nvidia might drop a cool $500 million on the UK’s Wayve, aiming to deepen its footprint in the self-driving car space, which ties into the bigger story of chipmakers strategically investing to boost their tech stack.

On the healthcare front, Delphi-2M is breaking new ground by predicting long-term disease risks using UK and Danish datasets, though it’s not quite ready for clinical use due to concerns about bias and privacy. Lastly, OpenAI just gave Codex a serious upgrade with GPT-5-Codex, making it a more reliable coding partner with smarter features for developers—could be a game-changer for those looking to streamline complex coding tasks.

Stories

Meta ships Ray‑Ban Display smart glasses (with a Neural Band) as a consumer AR+AI product

At Meta Connect, Meta unveiled the Meta Ray‑Ban Display — its first Ray‑Ban‑branded smart glasses with a built‑in color display for the right lens and an accompanying "Neural Band" wrist controller that uses surface electromyography (sEMG) for gesture input. Priced at $799 and slated to ship Sept. 30 in the U.S., the glasses deliver notifications, live captions, translations, navigation and short video playback while aiming to bridge Meta’s earlier audio‑only wearables and its more experimental Orion AR prototypes. Why it matters: this is a notable consumer push to put visible AR + on‑device AI into a familiar eyewear form factor, signaling Meta’s strategy to move AI from phone screens to always‑worn wearables — a move that could reshape how people access context‑aware AI services and accelerate app/partner ecosystems for AI glasses.
Read more → Reuters

Nvidia in talks to invest $500M in UK self‑driving startup Wayve

Nvidia has signed a letter of intent to evaluate a possible $500 million strategic investment in Wayve’s next funding round, the companies said. The move builds on an ongoing partnership (Nvidia participated in Wayve’s 2024 $1B+ round) and comes amid Nvidia’s broader pledge to pump billions into the UK AI ecosystem. Why it matters: a major strategic bet by Nvidia would deepen its ties to an end‑to‑end, ML‑centric autonomous driving approach (Wayve’s camera‑first, foundation‑model style stack), accelerate in‑vehicle adoption of Nvidia Drive compute, and underscore how chipmakers are using direct investments to secure both customers and technology roadmaps in the race to commercialize autonomy.
Read more → TechCrunch

Delphi-2M — a generative transformer that predicts risk for 1,000+ diseases up to decades ahead

Researchers from EMBL, the German Cancer Research Centre and collaborators published Delphi-2M, a generative-transformer model trained on anonymized UK Biobank records (~400k people) and externally validated on ~1.9M Danish registry records. The model forecasts individual risk trajectories for more than 1,000 ICD-coded conditions (and death) up to ~20 years ahead, with performance comparable to many single-disease risk tools for common conditions. Why it matters: Delphi-2M demonstrates how transformer-style generative models can be repurposed for large-scale, longitudinal clinical forecasting — enabling population-level health planning and more holistic, multi-disease risk profiling. Impact and caveats: the work could accelerate research into prevention and personalized care, but the authors and commentators caution it’s not yet a clinical diagnostic tool — questions remain about bias, generalizability across populations, privacy, and commercialization.
Read more → The Guardian

Atlassian to buy developer‑intelligence startup DX for about $1B to deepen AI insights for engineering teams

Atlassian announced an agreement to acquire DX, a developer‑intelligence platform, for roughly $1 billion in cash and restricted stock. The deal (expected to close in Atlassian’s FY2026 Q2) will fold DX’s analytics for engineering workflows, productivity and AI‑tool adoption into Atlassian’s suite (Jira, Bitbucket, etc.). Why it matters: as enterprises rush to deploy AI across engineering orgs, Atlassian is positioning itself to be the vendor that helps customers measure ROI and manage the people/process impacts of AI — a strategic move to lock in large enterprise customers and expand SaaS monetization around AI observability and governance.
Read more → Reuters

Check Point acquires AI‑native security firm Lakera to build end‑to‑end enterprise AI protection

Check Point announced a definitive agreement to acquire Lakera, an AI‑native security startup that provides pre‑deployment assessments and real‑time runtime enforcement for LLMs, agents and multimodal workflows. Financial terms were not disclosed; Check Point expects the deal to close in Q4 2025. Why it matters: with enterprise adoption of generative models and autonomous agents accelerating, vendors are consolidating AI security capabilities — Check Point’s move aims to embed runtime AI protections into its Infinity platform and position the company as a comprehensive vendor for securing the full AI lifecycle.
Read more → ITPro / ChannelPro (Future plc)

OpenAI upgrades Codex with GPT‑5‑Codex — a faster, more reliable coding partner

OpenAI rolled out GPT‑5‑Codex (a GPT‑5 variant) to its Codex products on Sept. 15, 2025 — optimizing the model for agentic coding tasks such as long refactors, code review and multi‑step engineering workflows. The upgrade promises higher benchmark performance, dynamic ‘thinking’ time (from seconds to hours depending on task complexity), better test-driven iteration, and transparency features (logs, citations, test results). It’s available across Codex tools (CLI, IDE extensions, Codex web, GitHub integrations and ChatGPT/Codex products) for paid plans — a meaningful step for developer productivity and the growing market for AI coding assistants.
Read more → TechCrunch

Mark Cuban Foundation opens 2025 free AI bootcamps — hands‑on AI education for high schoolers

The Mark Cuban Foundation announced its expanded 2025 AI Bootcamp program (applications open through Sept. 30, 2025), bringing free, hands‑on AI training to high school students and educators in dozens of U.S. cities. The three‑Saturday bootcamps teach generative AI basics, ethics, and project work with modern AI tools, plus a teacher fellowship to help scale AI curriculum — a practical, low‑cost pathway to learn AI and coding skills for underserved students.
Read more → GlobeNewswire / Mark Cuban Foundation