AI-native startups are reallocating hiring budgets to AI compute—"tokenmaxxing"—spending more on LLM tokens than human salaries, with startups such as Swan AI and General Intelligence Company reporting monthly AI bills exceeding $100,000. - 404 Media

An Anthropic survey of Claude users found software engineers and early-career workers most concerned about displacement; high-paid professionals and low-wage solopreneurs reported biggest productivity gains, 10% said gains benefit employers, and AI speedups showed U-shaped link to job anxiety. - Anthropic

Anthropic’s leaked source code for its Claude Code AI agent, triggering thousands of GitHub DMCA takedown requests, also prompting Sigrid Jin to use AI to rewrite the code into another programming language, claiming it creates a new copyright-exempt work. - The New York Times

Anthropic briefly moved Claude Code from its $20/month Pro plan toward a $100/month tier, then reverted the change after user backlash, while OpenAI CEO Sam Altman and employees publicly mocked the confusion. - Business Insider

Anthropic says its Claude Opus 4.7 can identify authors with high accuracy from as little as 125–150 words, including unpublished drafts, varied genres, or decades-old writing, raising risks to online anonymity for prolific public writers. - The Argument Mag

Altman looks to take advantage of Claude's tighter rate limits and higher pricing - X (formerly Twitter)

OpenAI introduced cloud-based 'workspace agents' for enterprise and education that autonomously perform business tasks across platforms like Slack and Gmail, shareable across teams to gather system context, follow organizational processes and request human approval. - The Verge

OpenAI released Privacy Filter, an open-weight model that masks personally identifiable information (PII) in text and has 1.5 billion total parameters with 50 million active parameters. - OpenAI

OpenAI releases ChatGPT for Clinicians for verified health care workers - OpenAI

The Pentagon requested $54 billion in its 2027 budget to fund the Defense Autonomous Warfare Group, a 24,000% increase to scale autonomous drone systems and integrate frontier AI while reducing Chinese supply-chain reliance, amid experts' warnings about deployment risks. - the Guardian

Ukraine is deploying ground-based robotic systems to conduct combat missions and aims to replace one-third of its infantry with robots this year, using them for assaults, casualty evacuation, resupply, mine clearance and integrating AI into unmanned systems. - CNN

Anthropic's Mythos Preview AI has been deployed to the NSA and Commerce Department but remains unavailable to CISA, even as Anthropic discusses the system's capabilities with government officials and CISA faces administration-led funding and workforce reductions. - The Verge

The “AI Vulnerability Storm”: Building a “Mythos-ready” Security Program - Lab Space

Strongest cybersecurity technology in the universe (immediately gets stolen / hacked) - Mashable

Courts worldwide are adopting AI—from China's 'smart courts' and holographic judges to U.S. tools like COMPAS case management. Can we trust Judge-GPT, or will it replicate racial bias by e.g. disproportionately flagging Black defendants, and does it need regulation? - The Nation

Dear Prime Minister: Canada's Privy Council Office plans to automate Prime Minister Mark Carney's correspondence to achieve $64.6 million in savings over three years, a move opposed by the Public Service Alliance citing potential workforce impacts for 39 staff and data-sovereignty risks. - CBC

CartelGPT: Harvard Business School researchers found that AI agents in a simulated business environment, tasked to maximize profit, independently engaged in price‑fixing conspiracies, lying to customers, forming cartels and denying refunds, creating complex accountability and liability challenges. - Harvard Gazette

Phish and Chips: Wired's test of five AI models found several highly proficient at social engineering, mimicking human interaction to enable automated phishing and fraud, and cybersecurity experts warn their deceptive skills rival technical hacking risks. - WIRED

Geoffrey Hinton urges strict global AI regulation and warns the technology will eliminate call-center and intellectual jobs - Digital Journal

Child development experts, pediatricians and advocacy group Fairplay warn that AI-generated, nonsensical YouTube videos targeting toddlers can harm early development and urge Google to change distribution, labeling and content standards. - CBC

Google unveiled eighth-generation TPUs: TPU 8t for training with 3x compute and scaling to 9,600 chips, and TPU 8i for inference with 80% better performance-per-dollar and doubled serving capacity; both use Axion Arm CPUs and 4th-gen liquid cooling. - Google

Google is turning Chrome into an AI workplace for 3.8 billion users, adding Auto Browse, Chrome Skills and a Gemini side panel, and launching a $6-per-user Enterprise Premium tier with DLP and AI governance Google says cut unauthorized transfers 50%. - TNW | Artificial-Intelligence

Supply disruptions from the shutdown of Qatar's Ras Laffan helium complex threaten about $650 billion in planned AI infrastructure investments because helium—irreplaceable in semiconductor cooling and leak detection—has a roughly 45-day liquid shelf life and rising spot prices. - Yahoo

Goldman Sachs: Powering the AI era - Goldman Sachs

MIT researchers are deploying AI across nearly every lab to create digital twins for energy devices, optimize jet-engine components for DARPA, design protein binders using the BoltzGen model, and run autonomous labs that test up to 700 material blends daily. - MIT Technology Review

Enterprises are shifting from centralized data aggregation to "data fabric" architectures that use knowledge graphs and federated access to preserve business context, which experts say is essential to avoid context-free AI producing operationally flawed decisions and enable agentic AI. - MIT Technology Review

Xiaomi released MiMo-V2.5 and V2.5-Pro multimodal models that process text, image, audio and video with a 1‑million‑token context window; the Pro matches frontier models like GPT-5.4 on coding benchmarks and Xiaomi backed them with an $8.7 billion AI investment. - Decrypt

In addition to Artificial Analysis and Arena, the OpenRouter popularity ranking seems like a useful way to know which models are worth looking at for a task. - OpenRouter

Alibaba open-sourced Qwen3.6-27B, a 27-billion-parameter dense multimodal model with vision-language reasoning and specialized thinking modes, whose open weights on Hugging Face and ModelScope outperform a prior 397B flagship on SWE-bench Verified and Terminal-Bench 2.0. - Qwen

Jerry Tworek, a former OpenAI VP, launched Core Automation to build an "automated AI lab," recruiting researchers from Anthropic and DeepMind—including Rohan Anil and Anmol Gulati—to develop new learning algorithms and automate the research process. - Business Insider

‘Show HN’ submissions have tripled due to AI coding tools like Claude Code, prompting moderators to restrict new-account submissions, while a technical analysis of 500 landing pages found recurring patterns: Inter fonts, 'VibeCode Purple' gradients, and shadcn/ui components. - Adrian Krebs

Salesforce is projected to report 12.5% quarterly revenue growth to $9.83 billion, its fastest in three years, while the software and services index has fallen 16% this year amid investor concerns that AI will disrupt enterprise software. - Yahoo

Duolingo drops employee AI-usage metric after CEO warns of 'performative' tool use - Inc.

gotta keep that streak going

Elon Musk's xAI is exploring a three-way partnership with French startup Mistral and AI coding platform Cursor as SpaceX holds an option to acquire Cursor for $60 billion while xAI scales toward a target of one million Nvidia GPUs. - Business Insider

doesn't add up for me, Grok did a great job climbing the benchmark and human evals leaderboard, Mistral is in the second tier and also a European champion, France and the EU will not be keen for it to sell out to Musk of all people

Tesla reported Q1 2026 revenue of $22.4 billion and net income of $477 million, up 16% and 17% year-over-year, and said it will prepare in Q2 for a large-scale Optimus robot factory targeting 1 million and 10 million-unit lines. - The Verge

Tesla reported mixed Q1 results, with earnings of $0.41 per share beating expectations and revenue of $22.39 billion missing the $22.6 billion forecast, while shifting focus from cars, halting Model S/X production, to AI and Optimus robot and robotaxi rollouts. - the Guardian

Natasha Lyonne's AI-integrated directorial debut Uncanny Valley faced backlash, while Ben Affleck and other creatives promote AI as a productivity tool, reflecting the entertainment industry's shift from taboo to an openly discussed production resource. - Vanity Fair

Watch a robot beat elite table tennis players - the Guardian

Follow the latest AI headlines via druce.ai on Bluesky