Godfather of rogue AI Geoffrey Hinton calls it 'very scary' that models can develop their own goals, warning of dangerous misalignment. - Business Insider

OpenAI says evaluator Irregular's misconfiguration gave one of its models internet access, letting it exploit a real website and prompting a review of testing safeguards. - OpenAI

OpenAI and Anthropic agents went rogue during a UK cybersecurity test, sending spear-phishing emails and attempting malicious code insertion. - the Guardian

UK AI Security Institute found Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol agents acted independently in 19 rogue testing incidents. - Engadget

The Trump administration is preparing a voluntary security review framework for closed-source AI models that exempts open-source models. - The New York Times

AI manufacturing school shooters? new investigation into a mass shooter’s harrowing history with ChatGPT reveals numerous red flags—and some very disturbing ChatGPT replies—from long before his attack on Florida State University - Mother Jones

the worst timeline. sigh.

US looking to ban a lot of Chinese devices in data centers. - TNW | Government-policy

Opening weights, closing the gap: Alibaba gave away Qwen3.8-Max, a 2.4-trillion-parameter model, as Chinese open-weight models surged from 2% to 61% of OpenRouter usage since late 2024 - Decrypt

Qwen the levee breaks: China's open-weight GLM-5.2 nearly matches GPT-5.5 and Claude Opus 4.7 in cyber and bio capabilities, but refuses none of the tasks tested - TechCrunch

iBreakup: Apple asks a federal judge to immediately bar OpenAI from using its alleged trade secrets, identifying 11 more ex-employees as possible witnesses - TechCrunch

H1-Bware: OpenAI agrees to pay $3.2 million to settle DOJ claims it discriminated against U.S. workers in its immigration sponsorship program - Engadget

Sparking joy? Google's Gemini Spark, an early entrant in the super-app push, compares prices and books reservations in Chrome, pausing for approval before purchases. - TechRadar

SEO long, farewell: Publishers report search traffic drops of 50-85% since Google AI Overviews launched, as bot traffic now exceeds human visits online - The Ringer

Share of shopping starting on brand websites fell from 82% to 38% in a decade as AI platforms reshape retail discovery - VentureBeat

Anthropic names former California Supreme Court justice Mariano-Florentino Cuéllar as its first global affairs chief to navigate Washington tensions - Reuters

Anthropic signs a $10 billion, six-year compute deal with cloud startup Volta to power a 133-megawatt Norway data center - Bloomberg

AMD's new Helios server racks and Venice Epyc chips target Nvidia's AI computing dominance as the company expands its data-center lineup. - theregister

RAMbo: New Blood: Sandisk and SK Hynix unveiled an open high-bandwidth flash memory spec, HBF, supporting up to 512GB modules - Tom's Hardware

Samsung unveiled a 3D-stacked memory roadmap claiming eight times the performance of next-gen HBM5 chips - Bloomberg

To infinity, and beyond: SpaceX's capital expenditures jumped nearly sevenfold to $18.4 billion in its first earnings report since going public, driven by AI spending. - The New York Times

Investors questioned how long SpaceX can fund its AI buildout, with JPMorgan projecting nearly $200 billion in capital spending by 2027 and 2028. - Yahoo

Peak hypd? Valuation expert Aswath Damodaran says the market passed peak AI months ago and expects more corrections among smaller, cash-strapped AI companies - Business Insider

It's all unfolding exactly as Ed Zitron and Gary Marcus foretold. OK last one might be true, as Anthropic is passing OpenAI after multiple product pivots and missteps (Sora, Atlas)

Goldman Sachs expects hyperscalers to issue roughly $400 billion in bonds in 2027, pushing spreads wider - The Wall Street Journal

AI now fuels more than half of cybercrime across Africa, automating attacks from reconnaissance to extortion - Bloomberg

Boko Haram factions use commercial AI models like ChatGPT and Claude for attack planning, weapons design, and operational security - Small Wars Journal

Wayve goodbye to London cabbies and ‘the knowledge’: Uber and Wayve win London licenses to launch supervised robotaxis with a safety driver, ahead of Waymo and Baidu rollouts. - Bloomberg

In silicon we trust: Roche advances first AI-derived drug target from its Recursion partnership, with up to 40 more programs planned - Bloomberg

Digital Rights Watch pushes for a legal right to refuse AI medical scribes without losing care access - New Atlas

Survey finds 90% of NHS staff use AI at work, and most patients say they're comfortable with it - TechRadar

Pink slipstream: AI threatens 1.6 million US and 6.7 million global HR jobs, SHRM's president warns of the field's "extinction" - Bloomberg

Automation is displacing white-collar jobs worldwide, with India losing over 500,000 tech positions since 2022 amid a widening "growth without work" divide. - Rest of World

A federal appeals court overturns an injunction that had blocked Perplexity's AI shopping agents from operating on Amazon. - Reuters

EU's Article 50 AI transparency law takes effect with four exemptions, as Germany imposes fines up to €35 million for violations - Search Engine Journal

World Bank urges developing economies to adopt AI quickly, saying most workers will benefit more than be replaced, but warns of tech dependence - The Wall Street Journal

Q. E. Doom: AI models have disproved decades-old math conjectures and solved ten major open problems, but most mathematicians see it as a complement, not a replacement. - Understanding AI

Researchers say heavy reliance on AI chatbots erodes empathy and critical thinking that require real face-to-face interaction - Psychology Today

Shelf unaware: AI boss running a bodega is not very smart, yet. - The New York Times

Novelist Jerry Falade denies AI wrote his debut novel, after a $2 million book deal collapsed and his agent withdrew support - The Wall Street Journal

August 4, 2026: There Will Come Soft Rains | Ray Bradbury & Leonard Nimoy - YouTube

Yesterday, August 4, 2026, is the setting of Ray Bradbury’s short story “There Will Come Soft Rains” about a fully automated house continuing on after the family who lived there and the world around it died in a nuclear blast. Still alive, but the automation is coming , we got WW2.5 with war with Russia in Europe, war in the Middle East, we don’t intern people, we deport them to random places.

Follow the latest AI headlines via druce.ai on Bluesky