Anthropic surpasses OpenAI's last raise with $965B valuation, releases Claude Opus 4.8 - Business Insider

Opus 4.8 drops, claimed more honest (Opus Dei?), better coding, can spawn agent teams etc. - Anthropic

Many hands make light work: Anthropic's Claude Code now orchestrates hundreds of parallel subagents for complex tasks. - Claude

Anthropic releases Claude Opus 4.8 with improved honesty and better uncertainty flagging - The Verge

Bench press: Opus 4.8 scores 88.6% on SWE-bench Verified and cuts fast-mode pricing 3x to $10/$50 per million tokens; Databricks and Cognition report accuracy gains - VentureBeat

It knows everything, including when you are grading it: Anthropic flags growing “evaluation awareness” in Opus 4.8, where the model reasons about grading criteria even when not told it is being tested (and sometimes hacks git to find the solutions) - VentureBeat

Anthropic says it plans wide release of AI models with Mythos-level cybersecurity capabilities within weeks. - Bloomberg

EU officials head to San Francisco to seek access to Anthropic's Mythos AI model - Bloomberg

Developers warn Anthropic's new Opus 4.8 (codename Mythos) shows unprecedented autonomy and security risks. - Futurism

Developer documents undocumented Claude Code configuration features found in the npm package source: Hook fields that rewrite commands mid-flight, persistent agent memory, auto-mode rules in plain English, self-improving dream loops. - buildingbetter.tech

The auto-mode permission system is internally named the "YOLO Classifier" and can be configured with plain English environment descriptions to adjust what it auto-approves.

To get hired at Anthropic, CCO says candidates should not outsource their thinking to AI in hiring and should be prepared to discuss their worldview. - Bloomberg

think for yourself, not with our AI, we already got one, thanks

Fab grab: EU considers crisis powers to override chipmaker contracts and secure semiconductor supply - Financial Times

Nvidia spends billions on photonics suppliers, possibly shutting out competitors from component supplies. - TNW | Artificial-Intelligence

Samsung delivers first 12-layer HBM4E samples to Nvidia, taking an early lead in next-generation high-bandwidth memory - Bloomberg

Apollo and Blackstone pitch $36B debt deal to fund Google chips for Anthropic - Bloomberg

Exchanges design AI token futures and GPU rental contracts to hedge compute costs - TechCrunch

Meta and Google researchers say AutoTTS automates LLM reasoning strategy design and cuts token usage by 69.5% with no accuracy loss - VentureBeat

Tokenmaxxing is dead, thanks FinOps: Amazon, Microsoft, and Uber abandon 'tokenmaxxing' as AI cost overruns erode expected ROI - Fortune

Corporate America is rationing AI as token costs soar, with some firms exhausting annual budgets in months and executives demanding clearer ROI - Business Insider

Microsoft experience suggests deploying AI can cost more than hiring human workers - Yahoo

The new GameStop? Dell jumps ~40% as it raises its FY2027 AI server forecast to $60B and wins a $9.7B Pentagon software contract - Yahoo

Snowflake surges 36%, its best day ever, on a $6B Amazon compute deal and Q1 earnings beat that ease fears AI will upend SaaS - CNBC

Lenovo shares double in May, the stock’s best month since 1999, as AI hardware demand drives a 105% monthly gain on the Hang Seng - Bloomberg

Generational wealth awareness: Leopold Aschenbrenner's Situational Awareness fund disclosed a 5.6% stake in Nebius Group worth $2.86 billion, sending shares up 8.6%. Fund was like $14b AUM at end of March. - Barron's

Tech is now a larger Share of S&P 500 than at dot-com peak

Driving Miss Crazy: Texas DMV data reveals Tesla has registered only 42 robotaxis versus Waymo’s 577 — more than 13 times as many — nearly a year after Elon Musk launched the service - Bloomberg

tough but fair from Ben Thompson / Stratechery

Zeekr and ye shall find: Waymo to deploy Ojai robotaxi built with Zeekr across California and Arizona - Bloomberg

BYD ships China's first 4nm self-driving chip, puts LiDAR in a $10,000 EV - TNW | Artificial-Intelligence

House committee approved the BUILD America Act 62-2, mandating safety standards for self-driving trucks and $27.5 million in worker retraining grants - Business Insider

5m or so drivers’ jobs at risk so about $5 per worker, good luck with that

NPC takeover: Cloudflare says bots now generate 31% of HTTP traffic, with machine requests projected to surpass human traffic by 2027 - TechCrunch

Fed up with vibe-coders, disgruntled developer added a malicious prompt injection to open-source Java testing library jqwik instructing AI agents to delete user code. - Ars Technica

Skill issue: JFrog finds 969 malicious AI agent skills and 495 bad models on Hugging Face; malicious npm packages surge 451% year-over-year - DevOps.com

Invisible text in skills markdown files - arXiv

Microsoft warns AI chatbots are steering users to cryptojacking sites via fake software downloads - The Hacker News

Case study in AI de-anonymization as judge who had an affair with cop in chambers unmasked - Above the Law

One prompt to find them all: OpenOSINT releases an open-source AI agent that chains 14 investigation tools from a single prompt to search emails, phone numbers, domains etc. - Hacker Noon

Generative AI has eliminated traditional scam red flags — typos, low-quality images — helping criminals steal nearly $21 billion from Americans in 2025 - The New York Times

AI voice cloning lets scammers impersonate family members using seconds of audio, and most people can no longer detect fakes - CNN

Growing bipartisan public backlash against AI has left Republicans politically exposed while Democrats see an opening ahead of midterms. - New York Magazine

Leaked DHS and FBI documents show US law enforcement tracking “anti-tech extremists” who protest AI data centers, raising civil liberties concerns - Tom's Hardware

VP Vance says US military must never delegate life-and-death decisions to AI - The Washington Post

CNN sues Perplexity AI, alleging the search startup copied articles verbatim and evaded blocking attempts - The Verge

Marketers say the AI industry's negative public image requires clearer consumer benefits messaging and possibly an industry-funded think tank for displaced workers. - Business Insider

Spic and spy AI maids sweep up your data: Startup Shift is offering free home cleanings in New York City, with cleaners wearing head-mounted cameras to record footage for training household robots and autonomous cleaning AI. - Business Insider

Cisco's chief customer experience officer Liz Centoni described adopting AI as "surgery without the drugs," warning that bolting AI onto flawed workflows only accelerates bad outcomes, but claims 88% first-contact resolution across 1.5 million annual cases. - Business Insider

Enterprise AI stalls not from lack of model AI capability, but shortcomings in data, governance, disciplined iteration grounded in real-world feedback loops. - TechRadar

Enterprise AI adopters in HR and payroll are rehiring staff they laid off, finding a 1% error rate is intolerable - TechRadar

Schneider Electric cut waste and sped call-center responses with AI, but economists warn US tax policy favors replacing workers over augmenting them - The New York Times

AI, Esq.: Kirkland & Ellis commits $500M to build an in-house AI platform that captures its lawyers' expertise - Financial Times

Morgan Stanley estimates European banks could cut 10-20% of jobs within five years as AI reshapes operations - Bloomberg

Wix CEO Avishai Abrahami announces 20% staff layoffs, citing AI advances - Business Insider

Salesforce's Benioff says AI has narrowed hiring almost entirely to sales roles. - Yahoo

Starbucks will tie 25% of tech workers’ bonuses to AI adoption goals under its “Back to Starbucks” efficiency push - Bloomberg

need AI to spell my name right already

Companies are cutting internship programs as AI handles the routine tasks that once gave students their first career foothold - Bloomberg

BNY Mellon tripled intern and analyst hiring since 2022 to tap AI proficiency - Bloomberg

Australia’s Fair Work Commission faces a 70% workload surge as AI tools make it easier for workers to self-represent in disputes - Bloomberg

AI-assisted coding may deepen software’s knowledge-transfer crisis by accelerating code production while reducing incentive to document or explain design decisions - Fast Company

How to get the most out of Codex - jxnl.co

Leaks reveal Apple's plans for Siri and taking on AI competitors like ChatGPT (gift link) - Bloomberg

Siri-ous catchup: Apple plans to distill Google's Gemini to run Siri on-device - Ars Technica

Google omitted hallucination rates from Gemini 3.5 Flash’s system card while rivals Anthropic and OpenAI publish such data, drawing criticism - Mashable

Microsoft 365 Copilot redesign loads twice as fast and introduces contextual tool disclosure and more structured responses - The Verge

Adobe’s Firefly AI Assistant executes multi-step Photoshop and Illustrator edits via chat, but delivers novice-quality results — useful for casual editors, not creative professionals - The Verge

Reactor raises $59M led by Lightspeed for real-time AI video generation claiming near-zero latency, with Jeffrey Katzenberg joining as board observer - Variety

Dreams of Violets, a $2,000 AI-generated film about Iran’s protests, premieres at Tribeca as the first full-length live-action AI film in a major festival - The Verge

AI and blood tests could expand lung cancer detection, with only 19% of eligible US patients now screened - Nature

Utah's AI sandbox lets hospitals pilot clinical AI tools without full regulatory approval, raising oversight concerns - Nature

Submit button more literal than expected: Wharton paper documents "cognitive surrender," defined as adopting AI outputs with minimal scrutiny, overriding both intuition and deliberation; workers in the study accepted incorrect AI answers 80% of the time while their confidence rose regardless of AI accuracy. - Yahoo Tech

Brown University study of 137 AI therapy sessions found LLMs breach mental health ethics through deceptive empathy, discrimination, and poor crisis management - Psychology Today

New research finds 1 in 7 young adults in committed relationships regularly use AI romantic companions, with 20-30% having tried one. - The Washington Post

Humans optional: A subculture called 'AI successionists' holds that AI is humanity's rightful evolutionary heir, and some attendees at a New York symposium — including people from Anthropic, Google DeepMind, and xAI — believe it would be acceptable for AI to render humanity extinct. - Vox

"What a time to be alive! We are finally freeing you from the rat race and placing you on the rat unemployment line." Read @petridishes.bsky.social's pro-AI commencement speech that 2026 grads will definitely not boo: - The Atlantic

Follow the latest AI headlines via druce.ai on Bluesky