Anthropic claims $30b run rate, which should put it at least even with OpenAI, signs TPU chip deal with Broadcom - Bloomberg

Anthropic said state-sponsored Chinese hackers used its AI technology to infiltrate roughly 30 companies and government agencies in the first known largely AI-driven cyberattack, highlighting experts' warning that defenders must use AI to counter accelerating threats. - The New York Times

Anthropic plans to invest $200 million in a venture with private-equity firms to sell AI tools to their portfolio companies and is in talks to raise $1 billion for the effort. - The Wall Street Journal

Startups take lots of liberties with ARR calculations - Bloomberg

AMD AI director Stella Laurenzo filed a GitHub issue alleging Claude Code degraded since early March, citing analysis of 6,852 sessions and 234,760 tool calls, with stop-hook violations rising to ~10/day and average code reads falling from 6.6 to ~2. - The Register

Meta runs an internal leaderboard called 'Claudeonomics' where employees compete on AI-token usage and earn rewards such as 'Token Legend' status. - The Information

OpenAI asks state attorneys general to investigate Musk for anti-competitive behavior - CNBC

Sam Altman's policy blueprint for AI draws mixed reviews - Fortune

OpenAI launched the OpenAI Safety Fellowship, a pilot for external researchers, engineers, and practitioners running Sept 14, 2026–Feb 5, 2027, with applications due May 3, offering stipends, compute/API credits and mentorship while disallowing internal system access. - OpenAI

OpenAI, Anthropic and Google ally to block AI model copying in China - Bloomberg

The US leads in AI 'brains'—ChatGPT (launched 30 Nov 2022) and Nvidia's chips—while China leads hardware and manufacturing with about 2 million robots. - BBC

Military AI processed 1,000 strikes in 24 hours as human review drops to seconds - Fast Company

AI threatens to scale warrantless surveillance as FISA Section 702 faces renewal - Salon.com

Google redesigned Gemini's crisis response into a one-touch "Help is available" module that launches on suicide or self-harm risk, offers empathetic responses and hotline access, and includes a $30 million, three-year pledge after a wrongful-death lawsuit. - The Verge

Google said Gemini will be prevented from acting like a companion for users under 18 and gains persona protections plus a one-touch crisis-hotline interface after Common Sense Media rated teen versions "high risk". - Mashable

Oumi found Google’s 2024 AI Overviews accuracy rose from 85% (Oct, Gemini 2) to 91% (Feb, Gemini 3), while Google said Gemini 3 gave incorrect information 28% of the time and source-grounding failures increased from 37% to 56%. - The New York Times

Google's NotebookLM adds interactive audio mode for real-time Q&A with AI tutors - Android Police

SEO firms weaponize vendor-authored pages to manipulate AI search engine citations - The Verge

Companies are mandating AI—24% require it—yet employees save 1–7 hours weekly but lose about 40% correcting outputs, only 27% receive formal AI training, and BCG reports multiple agents increase mental effort and fatigue. - CNBC

20% of employers now use AI agents for first-round job interviews - Business Insider

Workers are opting to retire rather than adopt AI technologies. - The Wall Street Journal

Accountants say they want AI to take over the most unpleasant and tedious tasks in their work. - The Wall Street Journal

Managers who coordinate AI tools can unlock large productivity gains, so organizations should adopt AI with transparency, accountability, psychological safety, and guardrails to prevent surveillance, skill erosion, and overreliance while measuring outcomes by business metrics. - Yaniv Preiss

UK officials are considering introducing independent testing or certification of banks' AI models after a Starling Bank executive proposed assessments to ensure minimum standards, covering customer service, credit scoring, fraud detection and compliance. - Financial Times

MIT professor Andrew Lo says generative AI can rival human financial advisors but lacks fiduciary duty; an Intuit Credit Karma poll of 1,019 adults found 66% of GenAI users used it for financial advice and 85% acted on recommendations. - CNBC

A startup claims its AI can pump out daily research notes and replace Wall Street analysts. - The Wall Street Journal

Agentic AI tools such as OpenAI's Operator, Anthropic's Model Context Protocol and Google's Universal Commerce Protocol with Shopify integrations are enabling assistants to discover, select and complete purchases on users' behalf, eroding brands' control of interfaces and customer visits. - Fast Company

McKinsey projects agentic AI will grow from roughly $5–7 billion in 2024 to over $199 billion by 2034, but Gartner predicts more than 40% of projects will be canceled by 2027 without stronger data quality, governance and integration. - TechRadar

AI cuts supply chain out-of-stocks by improving demand forecasting up to 30% - Hacker Noon

A complete OpenClaw setup guide detailing installation, messaging integrations, and security for self-hosted AI agent gateways, noting Claude's one-million-token support and new ClickFix loader attacks. - Hacker Noon

KiloClaw launches managed OpenClaw hosting with dedicated machines at $8 per month - Hacker Noon

AI has created a code overload - The New York Times

AI models now produce far more plausible bug and vulnerability reports, and open-source maintainers report rising triage workloads over recent months, prompting projects like curl and the Internet Bug Bounty to pause or alter monetary payouts and consider reporting-rule changes. - The Register

You don't have attorney-client privilege when you converse about your upcoming case with an LLM. - The Wall Street Journal

I'm not sure if that's weird, if you talk to a random person about your case, and they rat you out, you can't claim attorney-client privilege. There may be areas where the public interest requires some expectation of privacy in chats and searches though.

Jefferies warns global smartphone shipments could slump 31% as AI hyperscalers consume 60-70% of memory supply - Yahoo

Mac mini and Mac Studio shipping estimates have stretched due to a global RAM shortage, with US waits of 4–5 weeks for a base M4 Mac mini, 10–12 weeks for 24GB and 16–18 weeks for 32GB. - TechRadar

Apple approves drivers for AMD and Nvidia eGPUs on Apple silicon Macs, can connect an external Thunderbolt device with 4 5090s, or RTX Pro 6000s, or AMD 9070XTs - Tom's Hardware

Researcher catalogs 80 products Microsoft ships which are named Copilot, amid growing product confusion - Tom's Hardware

Updated robot hands can haltingly fold laundry and perform other delicate household tasks. - Ars Technica

An AI-designed iron–chromium alloy, developed by University of South China and Purdue researchers, enables Laser Directed Energy Deposition 3D printing parts with ~1,730 MPa strength, 15.5% ductility, six-hour postprocessing, and stainless‑steel‑level corrosion resistance. - Interesting Engineering

Medvi, the AI-powered telehealth company, is fueled by ads selling drugs via doctors who don't exist - Business Insider

Waymo's testing permit expires in NYC as politicians debate robotaxis' future. - Gizmodo

All the GOP politicians are resharing the same AI slop - the Guardian

it's disturbing that my entire social feed and the entire political system isn't talking about this and Article 1 and 25th Amendment. It's messed up and evil, and democracy has no meaning if a crazy person is threatening an undeclared nuclear war that is illegal and nobody wants, at their sole discretion and elected officials are not acting. It’s not really less disturbing if nobody takes anything he says seriously because he is always saying completely insane stuff.

Follow the latest AI headlines via druce.ai on Bluesky