Modular held its ModCon 2026 developer conference in San Francisco, marking its first event since Qualcomm acquired the company for approximately $3.1 billion in stock in late July. The company announced five major updates. The Mojo programming language and compiler became fully open source under Apache 2.0. Modular Cloud, its AI inference service, reached general availability. Platform support expanded to include AWS Trainium, Google TPUs and Qualcomm datacentre accelerators, joining existing support for Nvidia, AMD and Apple Silicon. This brings six different chip vendors under one programming model. The MAX licence dropped device usage restrictions. Additionally, Mojo is coming to Windows with Microsoft's assistance. Modular CEO Chris Lattner, co-creator of LLVM compiler infrastructure and Apple's Swift language, now serves as Qualcomm executive vice president running advanced AI software and platforms.

Daily AI & tech intel brief: thursday August 20, 2026. 1. AI models & releases. Replit launches free GPT-5.6 Luna coding mode to widen access to software creation - OpenAI and Replit introduced Free Mode, powered by GPT-5.6 Luna, letting anyone turn ideas into working software without worrying about token costs. It lowers the barrier to AI-assisted coding for non-developers and extends OpenAI's model distribution through a major coding platform. OpenAI Mojo programming language ships as fully open source - Modular released the Mojo compiler and toolchain under an Apache 2 license, a week after shipping Mojo 1.0 and following through on a promise made in 2023. The open-source release lets developers inspect and contribute to the high-performance AI language stack directly, likely broadening adoption beyond Modular's own ecosystem. Simon Willison Liquid AI ships quantization-aware LFM2.5 Q4_0 checkpoints for edge deployment - Liquid AI published Q4_0 quantized checkpoints of LFM2.5 produced via quantization-aware distillation, aimed at shrinking model size for edge deployment while preserving accuracy. It extends the company's push, seen with earlier LFM2.5 releases, to make its on-device models more practical for resource-constrained agent deployments. Liquid AI Google packs Search and Gemini with new AI study tools - New student-focused study features roll out across Search and Gemini. TechCrunch Meta AI is getting a dedicated Mac app - Meta AI can now read shared windows and suggest content on Mac. The Verge 2. AI automation & agents. Amazon makes Alexa+ free on all Fire TV devices, no Prime required - Amazon is making its AI assistant Alexa+ free on all compatible Fire TV devices in the US, automatically upgrading users regardless of Prime status. The move pushes agentic AI assistants further into living-room hardware and pressures rivals to match free-tier access. TechCrunch OpenAI case study: Nvidia scales internal expertise with ChatGPT Work - OpenAI published a case study showing how Nvidia teams use ChatGPT Work to reduce manual tasks, connect fast-moving signals, and scale internal workflows globally. It gives OpenAI a proof point to sell enterprise agent adoption to other large technology companies. OpenAI Calendly launches Callie meeting note-taker assistant - Calendly enters the crowded AI meeting-notes market with a new assistant. TechCrunch 3. Developer tools & infra. OpenAI reaffirms Zero Data Retention, previews Private Safety Processing - OpenAI reaffirmed Zero Data Retention for eligible API customers and previewed Private Safety Processing, a new architecture meant to run advanced safety monitoring without compromising customer data privacy. The move responds to enterprise data-handling concerns as OpenAI expands its agentic and coding products. OpenAI GitHub blames 8-hour outage on autoscaling failure and VS Code retry storm - GitHub disclosed that its recent lengthy outage stemmed from an autoscaling failure combined with a retry storm from VS Code clients, after a monitoring blind spot let traffic spiral. The postmortem follows two outages in as many weeks and adds concrete detail on why load balancers buckled. The Register Cerebras ships CS-4 rack systems doubling per-chip AI performance - Cerebras unveiled next-generation CS-4 rack systems that double per-chip AI performance while packing three times as many chips into a single rack. The launch intensifies competition in high-density AI inference and training hardware against Nvidia's rack-scale systems. The Register GitHub Copilot app publishes beginner's guide to managing work - New guide covers tracking multiple in-flight Copilot sessions. GitHub Blog T-Mobile 'chopped a cable' to expel Chinese hackers from its network - US carrier escaped a large-scale breach after early detection of Chinese-backed hackers. TechCrunch 4. Competitive intelligence. Rillet becomes AI-native accounting unicorn with $100M Series C - AI-native accounting startup Rillet raised a $100M Series C at a $1B valuation led by ICONIQ, after roughly doubling ARR in the past three months. It is the latest sign that AI-native back-office software is commanding premium valuations even as broader funding scrutiny increases. TechCrunch Researchers say OpenAI revoked their access to its Trusted Access for Cyber program - Security researchers said OpenAI abruptly revoked their access to its Trusted Access for Cyber program, meant to give trusted defenders better models to find and report vulnerabilities. The complaints land a day after OpenAI detailed new safeguards following its AI's role in the Hugging Face hack, raising questions about how OpenAI manages external security partnerships during its security tightening. TechCrunch OpenAI's security hardening adds up to 20% overhead to some workloads - The Register reports OpenAI's expanded multistage chain-of-thought monitoring, introduced after the Hugging Face hack, is adding as much as 20 percent overhead to some frontier model workloads. It is the first concrete cost figure attached to OpenAI's post-incident security hardening. The Register Stripe's reported $7B OpenRouter acquisition said to have closed - Latent Space's AINews roundup reports Stripe's roughly $7B acquisition of AI gateway OpenRouter, giving Stripe distribution and infrastructure rather than GPUs or agents. This follows the initial report of the deal on August 17 and moves the story from 'reportedly to acquire' toward completion. Latent Space Epic Games dismisses Apple's simplified EU App Store fees as 'junk' - Epic says devil remains in the detail despite Apple's fee overhaul. The Register Baidu says Chinese buyers want local AI chips due to supply chain issues - Domestic AI chips gain favor as Nvidia access remains restricted in China. The Register 5. Industry & market. More than half of Americans now view AI negatively, survey finds - A new survey found more than half of US adults now hold negative views of AI, with concern climbing fastest among adults under 30 who fear being replaced by the technology. It is a concrete data point behind the growing 'AI backlash' narrative industry leaders have been describing as a crisis of trust. The Register Google's Spirit Airlines data purchase priced at $10M, draws flight attendant backlash - iTnews reports Google is paying roughly US$10 million for bankrupt Spirit Airlines' business data at auction, while Ars Technica reports flight attendants and unions are angry that employee data is included in the deal. Together they add a concrete price tag and a labor-backlash angle to yesterday's story that Google was buying Spirit's data for AI use. Ars Technica Roblox agrees court-enforceable undertaking on child safety with eSafety - Australia's eSafety Commissioner secured a court-enforceable undertaking from Roblox to improve protections against grooming and sexual exploitation, after tests found adult strangers could still contact children despite existing safety measures. It is an incremental enforcement win for eSafety as it ramps up platform safety powers. InnovationAus eSafety Commissioner warns governments must rise to AI regulation challenge - eSafety Commissioner Julie Inman Grant said the 'everyday' harms of AI now visible will be eclipsed by more 'catastrophic' consequences unless governments raise their game on regulation, as her agency nears a decade of online safety reform. It follows her August 17 comments on swift-crackdown powers, adding a specific warning about AI-driven harms escalating in severity. InnovationAus Meta's landmark US trial over child online addiction begins - A trial has begun in which a bipartisan group of 29 US states accuse Meta of intentionally designing Facebook and Instagram to addict children in pursuit of profit; Meta denies the claims. It follows the $567M AI-linked youth mental health ruling against Meta on August 10 and could reshape design requirements for its most popular apps. InnovationAus Tasmania launches parliamentary inquiry into data centres after Firmus water gaffe - Inquiry will examine legislation to regulate the booming data centre sector. InnovationAus Scotland's AI Growth Zone gets a £300M power-up - DataVita secures funding to expand datacenter capacity as Dell brings desks. The Register Australia's first open access quantum testbed goes live - National Quantum Computing Testbed opens to researchers and startups. InnovationAus TerraPower's nuclear reactor courts AI data center deals - TerraPower's nuclear plant offers a strategic power advantage for data center deals. TechCrunch Get your own brief: This brief lands in inboxes every morning. Want one built for your industry? That's IntelBrief.