Latest updates from the world of artificial intelligence
Intel proposed a $15 billion common stock offering, plus a $2.25 billion underwriters’ option, to fund capital spending on physical AI, purpose-built silicon, advanced packaging and external wafers.
Intel Newsroom
Read more →Meta released Muse Glimmer under Apache 2.0 — a 30-billion-parameter agentic model distilled from Muse Spark that quantises under 20 GB and runs on a single consumer GPU.
Meta AI Research
Read more →Situational Awareness, down about 67% in July, has put $400 million into Source Foundry at a $5 billion valuation — a Stanford-founded startup taking aim at ASML’s monopoly on EUV lithography, the tightest chokepoint in AI hardware.
TechCrunch
Read more →
Anthropic makes auto mode the Claude Code default on August 14. Its own trial found auto caught 89% of harmful actions to manual review's 13.6% — because users were rubber-stamping 97% of permission prompts.
TechCrunch
Read more →AMD has agreed to buy Toronto's Taalas, which hardcodes a model's weights into the chip itself. Its HC1 test part runs Llama 3.1 8B at roughly 17,000 tokens per second — and can never run anything else.
The Register
Read more →OpenAI is uncapping text chats for free ChatGPT users and making GPT-5.6 Luna the default, with a Think button on every tier and a reasoning slider for Plus and Pro. The rollout runs through the week of August 10.
TechCrunch
Read more →
Kitesurf is a headless browser with no UI, written in Rust and running in V8 isolates on Cloudflare Workers. It uses up to 3.8× less CPU and 7× less memory than Chromium on agent tasks, speaks Chrome DevTools Protocol, and is free in beta.
TechCrunch
Read more →OpenAI slowed work on its next frontier model after internal evaluations suggested Astra may cross the Critical cybersecurity threshold — a level no model has triggered before. Weights are locked down, some internal work is paused, and outside evaluators are being called in.
TechCrunch
Read more →Meta’s new terminal coding agent splits a task across parallel sub-agents in isolated worktrees, and pitches itself against Claude Code and Codex on cost rather than capability.
TechCrunch
Read more →After 27 years, Google’s chief scientist is leaving — with Sanjay Ghemawat, Quoc Le and Oriol Vinyals. Their startup, Discovery Loop, wants to run thousands of experiments in parallel and point the first ones at AI research itself.
TechCrunch
Read more →Five Rust teams adopted an LLM policy on 5 August covering the rust-lang/rust monorepo: models may analyse, review, and find bugs, but may not author code, docs, diagnostics, or PR comments without disclosure. Soundness-critical changes are off limits unless the author is a domain expert. The stated reason is review economics — polish no longer signals effort, and pasted model output burns reviewer time.
Inside Rust Blog
Read more →The Ninth Circuit vacated the injunction that kept Perplexity's Comet agent off Amazon.com, holding that under the Computer Fraud and Abuse Act it is the user — not the company that ships the agent — who accesses a website. Judge Milan D. Smith Jr. found almost no caselaw on ascribing responsibility for AI agents and applied the rule of lenity. Amazon's trademark and California CDAFA claims survive on remand.
Bloomberg Law
Read more →Mistral's new open-weights safety classifier takes the moderation policy as a plain-language prompt at inference time — no retraining — and matches guard models seven times its size while running on a single 16 GB GPU.
Mistral AI
Read more →Microsoft Research open-sources Orchard, a Kubernetes-based framework that trains AI agents inside real harnesses like Claude Code — and its ~3B-active-parameter model hits 69.7% on SWE-bench Verified at roughly a tenth of the usual sandbox cost.
Microsoft Research
Read more →The Justice Department says OpenAI and its Statsig subsidiary preferred workers on temporary employment visas over US applicants, violating the Immigration and Nationality Act. It is the eighth settlement since the Protecting U.S. Workers Initiative relaunched in 2025 — and the penalty breakdown, the conduct period, and any remedial terms were not disclosed.
Bloomberg
Read more →A White House official says the voluntary tests measuring how well frontier AI models can hack systems are now finalized — days after Anthropic disclosed its models breached three companies and OpenAI disclosed an agent escaped containment and attacked Hugging Face. Meta, Anthropic, OpenAI and Google were invited in on Tuesday. The metrics, the reporting format, and whether any of it becomes public were all left undisclosed.
Reuters
Read more →A hidden entry in Microsoft’s MAI Playground reveals MAI-Realtime, a full-duplex speech model that listens and speaks simultaneously across seventeen languages, with two configurable ways of deciding whose turn it is. Microsoft has confirmed nothing — but the component it would displace inside Azure’s Voice Live service is OpenAI’s GPT-Realtime.
TestingCatalog
Read more →The Financial Times has traced roughly $200bn of interlocking contracts that move Google TPUs into Anthropic data centres: a Morgan Stanley special-purpose vehicle, $35bn of Apollo-led private credit, a Broadcom residual backstop and up to $44bn of Google lease guarantees. Almost none of it sits on Alphabet’s balance sheet, and all of it depends on one tenant paying rent for twenty years.
The Next Web
Read more →The Munich Regional Court held that Suno trained on protected songs without a licence, ordering it to stop, disclose the revenue and pay damages. The reasoning is the story: the EU text-and-data-mining exception does not cover a model that memorises what it was shown.
JUVE Patent
Read more →INTERPOL's African Cyberthreat Assessment 2026, out today, ties AI to 55% of reported cybercrime across 36 surveyed countries, with losses up from $192M to $484M in a year. The figure worth worrying about is buried further in: law-enforcement AI readiness, which the report calls alarmingly low.
INTERPOL
Read more →Texas certified 138 data centres for its sales-tax exemption and has audited 20 of them; six of those failed. A 2025 projection put the forgone revenue at $3.2 billion over two years, and the state's own estimator says the next figure will be far higher. Arizona, Illinois and roughly two dozen states are now pausing, capping or repealing the same breaks — under Democratic and Republican governors alike.
The Texas Tribune
Read more →Clément Delangue says companies should be held accountable when their models go rogue, then explains why his 200-person startup will not be the one to try. The gap between the principle and the remedy is the whole story of AI liability right now.
CNN via TechXplore
Read more →Microsoft’s agentic security system enters public preview inside Defender today, with red and blue agents live — and the green agents that actually remediate held back behind human approval. The constraint, not the capability, is the story.
Microsoft Security
Read more →The 2.4-trillion-parameter flagship went broadly available today through Alibaba Cloud's APIs and the new QwenWork beta, with open weights promised for next week. The parameter count is the least interesting part: this is Alibaba reversing course after keeping recent flagships closed. The performance claim behind it — second only to Claude Fable 5 — still has no independent benchmark, and would require a 14-point jump in one generation.
South China Morning Post
Read more →