MechaHitler, Esq., CFA: The AI that spawned MechaHitler and deepfake porn wants to be your legal advisor and Excel jockey - theregister

Token efficient, reality deficient? SpaceXAI and Cursor unveiled Grok 4.5, claiming 2x rivals' token efficiency, priced at $2/$6 per million tokens, under their $60 billion Cursor acquisition deal. - x.ai

Opus Dei: Musk calls Grok 4.5 an 'Opus-class' model, but benchmarks show it trailing Anthropic's Fable 5 and OpenAI's GPT-5.5, roughly on par with Opus 4.8. - TechCrunch

An expanded class action suit accuses X “of shielding child predators by obstructing police investigations into Grok-generated child sex abuse materials (CSAM).” - Ars Technica

A build-off pitted Grok 4.5 against GPT-5.5, Claude Opus 4.8 and Fable 5 on identical app-building prompts. - TryAI

OpenAI makes ChatGPT better at banter: With GPT-Live, talking, listening, and formulating answers all happen at once. - theregister

OpenAI unveils full-duplex GPT-Live voice models that listen and speak simultaneously, replacing Advanced Voice Mode. - OpenAI

SWE dreams: OpenAI says SWE-Bench Pro is broken, don't use it. - OpenAI

Khosla Ventures founder Vinod Khosla says his OpenAI bet could return 1,000x or more, calling his approach option-value investing. - The Wall Street Journal

Anthropic hits $1.2 trillion valuation on secondary markets, a 550% year-over-year increase, with shares scarce. - Business Insider

Anthropic's new "J-lens" reveals a silent workspace inside Claude that mirrors a leading theory of consciousness - VentureBeat

Anthropic interpretability research finds a Claude 'workspace' (akin to short-term memory) where concepts form before words, and swapping concepts changes its reasoning. - Psychology Today

Anthropic and AE Studio built GRAM, a training method that routes dangerous dual-use knowledge into removable modules without degrading overall model performance. - Anthropic

Alberta Einsteins: Meta will build its first Canadian data center, a $9 billion, 1-gigawatt facility in Alberta, taking two to three years to complete. - CNBC

Meta plans to start AI chip production in September, double compute capacity - Reuters

Global datacentres are set to grow 3.5x at a $7 trillion cost, with critics warning AI's power and water use need scrutiny. - the Guardian

Ray-Banhammer: Ray-Ban smart glasses flaw let modders disable the recording-indicator light; Meta is patching it and banning sellers of tampering tools. - TechRadar

Muse Image can search the web and write code to fulfill prompts, using "deliberate reasoning" before generating images, with a video model coming next. - SiliconANGLE

Talent agency CAA and union SAG-AFTRA criticized Meta for making Muse Image reuse opt-in by default, warning it threatens creators' control over their likeness. - Los Angeles Times

Samsung, SK Hynix and Micron shares fall over 20% into bear-market territory even after Samsung posted record profit. - Business Insider

Nvidia is pairing its GPUs with d-Matrix's Corsair inference chips, which d-Matrix says run tokens 10 times faster at a third of the cost. - TNW | Artificial-Intelligence

SambaNova's platform, which mixes Nvidia H200 GPUs with SN50 chips, hit 763 tokens per second running MiniMax M2.7 in third-party benchmarks. - theregister

China will let Alibaba, ByteDance and DeepSeek buy limited Nvidia H200 chips - Bloomberg

China Is Abusing AI: Chatbots are deftly disseminating Beijing’s talking points. - The Atlantic

Corporate America is starting to balk at the cost of AI agents. A cheap alternative from China looks more tempting than ever. - The Atlantic

Beijing is weighing tighter controls on domestic AI models as Chinese systems like DeepSeek gain traction among US companies. - The Wall Street Journal

Cognition releases SWE-1.7, a coding model it says nears GPT-5.5 and Opus-level intelligence at lower cost, available now in Devin. - Cognition

AI-driven application surges have outpaced recruiter tools, killing the traditional cover letter, says Juicebox CEO David Paffenholz. - Fast Company

Intern-ship has hit an iceberg: Entry-level job postings fell 29% since January 2024 as AI automates the tasks that once trained junior employees. - Hacker Noon

Have your bot call my bot: 63% of US job seekers were interviewed by AI in the past year, as Coinbase's bot screened 1.5 million applicants. - Business Insider

AI-titled job postings have tripled since 2022 to 1 in 12 listings, with most now outside tech roles, Indeed finds. - Fast Company

Staffing firms pivot to scarce, specialized AI roles as demand for top AI positions jumps 226%, even as tech cuts AI jobs. - TNW | Future-Of-Work

Big tech CEOs pivot away from warnings of an AI jobs "wipeout" for a productivity-boosting message, as public sentiment toward AI's workplace impact sours. - The Wall Street Journal

Banks are replacing operations and analyst roles with AI, as JPMorgan support staff fell 4% and Standard Chartered plans 7,800 cuts by 2030. - American Banker

Companies are recruiting enthusiastic employees as internal "AI champions" to win over skeptical, distrustful colleagues after heavy tech investment. - The Wall Street Journal

Nvidia's Jensen Huang says his engineers now prefer building AI agents to writing code, calling it a promotion rather than a job threat. - TNW | Nvidia

Five Eyes agencies warned AI can autonomously hack and deploy ransomware, and a Schneier essay argues unskilled actors will increasingly gain that capability. - Schneier on Security

NHTSA warns self-driving car makers after Waymo vehicles blocked ambulances and fire trucks, including a two-minute ambulance block during an Austin mass shooting. - Engadget

Waymo will drop human safety supervisors in San Diego, Las Vegas, Tampa, and Denver, expanding fully driverless service to four more cities. - Engadget

New Jersey's proposed three-year robotaxi pilot would require lidar and radar alongside cameras on robotaxis, effectively excluding Tesla's camera-only Robotaxi service. - The Verge

Databricks found coding-agent harness design cuts cost more than model choice, with GLM 5.2 matching Opus 4.8 at $1.28 per task versus $1.94. - Databricks

AI agents that work, retry, and coordinate on their own can quietly compound errors, and regulators now expect governance built into the loop itself. - Yahoo

After Anthropic restricted access to Fable 5, Prime Intellect's Vincent Weisser and Adaption's Sara Hooker say DIY self-improving loops can democratize AI. - WIRED

Lyzr used its own AI agent to run outreach for a $100 million Series B raise. - Bloomberg

A deep dive into robots: useful robots good at a few things might start happening soon, generalized humanoid robots are a long way off. - Ars Technica

Path Robotics said AI welding robots cut labor 91%, as Boston Dynamics and Apptronik showed humanoids moving from demos to deployment. - SiliconANGLE

Weave Robotics, founded by ex-Apple engineers, will ship its $7,999 Isaac 1 laundry-folding robot this fall, though it still needs occasional remote teleoperation. - Business Insider

Killer robots and just wars - The Washington Post

NATO builds an AI-driven 'Kill Web' of sensors, drones and satellites to detect and block Russian attacks on its eastern flank. - Business Insider

A journalist was wrongly detained by Minnesota police after Flock license plate cameras misread a data-entry error as a stolen vehicle. - The Drive

EU Parliament approves "Chat Control 1.0," allowing suspicionless scanning of private messages on Instagram, Discord, Snapchat, Gmail, and iCloud until 2028. - Patrick Breyer

Nobel laureate Omar Yaghi leaves UC Berkeley for full-time role leading Tsinghua University's new AI materials discovery institute, had criticized US science cuts. - Nature

Starbucks is developing AI-powered in-house tools to replace some Microsoft inventory and IBM maintenance software, per an internal presentation reviewed by Bloomberg. - Bloomberg

Microsoft's Brad Smith says the Trump administration's AI policy amounts to "regulation without transparent or complete rules." - Yahoo

Like a good neighbor, AI can screw up: State Farm's new AI tool gives agents 'instant summaries' of customers, but early errors have fueled pushback from over 1,000 readers. - The Wall Street Journal

Alpha School, a $65,000-a-year private school using AI tutors instead of teachers, expands to Santa Monica. - New York Post

Brown economics professor says most students used AI on take-home exam. - Inside Higher Ed

Half of parents worry their kids rely too much on AI, and only 22% see school-approved tools. - TNW | Artificial-Intelligence

Paris startup ZML, backed by Yann LeCun, released free software that runs AI models across Nvidia, AMD, Google TPU, Intel and Apple chips. - TNW | Artificial-Intelligence

Follow the latest AI headlines via druce.ai on Bluesky