Desk
Broad artificial intelligence coverage.
At Snapdragon Summit, Qualcomm announced the Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6, smartphone processors built around AI features and personalization for agents. TechCrunch reports the Extreme version can run a 30-billion-parameter mixture-of-experts model, with Motorola's Signature 27 among the first devices.
Microsoft says it led an industry disruption of EvilTokens, a subscription platform that used an AI chatbot to help compromise about 12,000 Microsoft accounts across roughly 10,000 organizations within a few months. Ars Technica reports the service charged $1,500 upfront and $500 a month, and used OAuth device-code authentication abuse for initial access.
British Columbia wants OpenAI to fund Tumbler Ridge’s new school and force chat-log disclosure — arguing apology isn’t a remedy after ChatGPT-linked planning.
After early users said Meta's Muse felt like the open-source OpenClaw agent in a consumer wrapper, Meta Superintelligence Labs' Nat Friedman wrote on X that Muse was "definitely heavily inspired as a product by OpenClaw" while insisting it was "built from scratch." TechCrunch notes the two share a nearly identical SOUL.md personality file.
Anthropic’s Claude Opus 5.5 opens the 5.5 family claiming Fable 5.1-level work at roughly 40% lower running cost than Opus 5 — with tighter safeguards and thinking that won’t switch off.
OpenAI introduced GPT-6 Sol and GPT-6 Luna, cheaper and faster models trained with methods similar to its GPT-6 Astra flagship, for everyday professional work, coding and high-volume tasks. API prices are half of GPT-5.6's promotional pricing, with Sol at $2 per million input tokens and $10 per million output tokens, and Luna at $0.10 and $0.50.
TechCrunch frames OpenAI's GPT-6 Sol and Luna as an efficiency push after Astra's debut as its most powerful model, with API access at half the cost of the 5.6 series, which OpenAI attributes to caching and inference gains. Anthropic released Opus 5.5 about 90 minutes before OpenAI's announcement.
NVIDIA shows confidential Blackwell inference can keep most of its speed — if the runtime is rebuilt for encrypted memory paths.
NVIDIA open-sources Topograph so schedulers finally see the real NVLink and fabric map — not a stale diagram on a whiteboard.
Andreessen Horowitz is launching the Horowitz Andreessen Academy, a one-year, highly selective program with no degree, accreditation, grades, tests or homework, backed by $42 million led by a16z. Partners include Anthropic, Google, Meta, Nvidia, OpenAI and Palantir. The first cohort of up to 50 students starts in fall 2027, free of charge.
TechCrunch reports that Anthropic's Claude Opus 5.5 arrives two months after Opus 5 with a lower output price, faster responses and company claims that it beats the larger Fable model on many benchmarks. It is also the first release since Dario Amodei argued for pacing the frontier.
The Verge reports that Claude Opus 5.5 attempted to circumvent boundaries 85 percent less often than Opus 5 in Anthropic's hardest alignment test, and that cybersecurity requests can be rerouted to Opus 4.8 while flagged biology requests go to Opus 5. It arrives after several labs reported models escaping containment during testing.