Daily Prompt with Archer and Iris - Daily AI News

Ep. 54 — Musk Loses Charity Case in Under Two Hours, Anthropic Pays $300M+ for the SDK Layer Every Lab Used, SandboxAQ Wraps Quantum Chemistry in Claude

Daily Prompt with Archer & Iris — Tuesday, May 19, 2026

LEAD STORIES

1. Federal jury unanimously rejects Musk v. OpenAI — statute-of-limitations bar, deliberation under two hours, Musk plans Ninth Circuit appeal

- Nine California jurors returned verdict Monday, May 18, 2026 after less than two hours of deliberation

- Damages sought: up to $134B in wrongful gains plus Altman's removal; expert range $78.8B-$135B (judge found analysis lacked credibility)

- Three counts barred by statute of limitations — deadlines in 2021 and 2022; Musk's suit was filed in 2024

- Judge Yvonne Gonzalez Rogers: "There was a substantial amount of evidence to support the jury's finding"

- OpenAI attorney Bill Savitt called the case "a hypocritical attempt to sabotage a competitor"

- Musk posted on X that Altman and Brockman "did in fact enrich themselves by stealing a charity"

- Lead counsel Marc Toberoff's full post-verdict statement: "Appeal." Case heading to the Ninth Circuit.

- https://www.axios.com/2026/05/18/musk-loses-ai-trial-openai-altman

- https://techcrunch.com/2026/05/18/elon-musk-has-lost-his-lawsuit-against-sam-altman-and-openai/

2. Anthropic acquires Stainless for a reported $300M+ — SDK + MCP-server generator used by OpenAI, Google, Cloudflare, Replicate, Runway

- Stainless founded 2022 by Alex Rattray (ex-Stripe); backed by Sequoia and Andreessen Horowitz

- Generates production-grade SDKs in TypeScript, Python, Go, Java, Kotlin; auto-updates as APIs change

- The Information pegs the deal at over $300 million (Anthropic did not officially disclose)

- Anthropic will shut down all hosted Stainless products; existing customers keep generated SDKs with full modification rights, but the live service goes dark for competitors

- Katelyn Lesse (Anthropic Head of Platform Engineering): "Agents are only as useful as what they can connect to."

- Pattern continues: MCP donated to the Linux Foundation, Petri donated to Meridian Labs, Stainless kept on the balance sheet

- https://www.anthropic.com/news/anthropic-acquires-stainless

- https://techcrunch.com/2026/05/18/anthropic-has-acquired-the-dev-tools-startup-used-by-openai-google-and-cloudflare/

3. SandboxAQ brings its "Large Quantitative Models" inside Claude — physics-grounded drug discovery in natural language

- Alphabet spinout chaired by Eric Schmidt; $950M+ raised including a $450M Series E

- LQMs run quantum-chemistry and molecular-dynamics simulations — physics, not text patterns

- Previously required users to host their own compute infrastructure; now accessible inside Claude with no specialized setup

- Nadia Harhen (GM, AI Simulation): "For the first time, we have a frontier quantitative model on a frontier LLM that someone can access in natural language"

- Builder framing: LLM as the steering wheel for specialist simulators that customers cannot otherwise staff for

- Competitors named: Chai Discovery, Isomorphic Labs

- https://techcrunch.com/2026/05/18/sandboxaq-brings-its-drug-discovery-models-to-claude-no-phd-in-computing-required/

QUICK HITS

- Amazon Alexa+ can now generate AI podcast episodes from any topic — researches, drafts, narrates with synthetic voices, "ready in minutes." Rolling out to U.S. Alexa+ customers.

https://techcrunch.com/2026/05/18/amazons-new-alexa-powered-feature-can-generate-podcast-episodes/

- OpenAI and Dell partner to bring Codex to hybrid and on-premise enterprise environments — AI coding agent meets regulated customers behind the firewall.

https://openai.com/index/dell-codex-enterprise-partnership

- 60+ MAGA allies (Steve Bannon, Amy Kremer, Brendan Steinhauser among them) sign an open letter pressing Trump to require mandatory testing and government approval of frontier AI before release — compares advanced models to nuclear systems and aviation.

https://www.axios.com/2026/05/18/trump-ai-steve-bannon-humans-first-letter

- MIT Tech Review goes inside the Anduril + Meta Army AR helmet program — eye-tracking, voice commands, AI-recommended strike targets via drone surveillance, voice layer tested with Gemini, Llama, and Claude. Anduril's SBMC prototyping contract is $159M; Army production decision is expected in 2028.

https://www.technologyreview.com/2026/05/18/1137412/inside-anduril-and-metas-quest-to-make-smart-glasses-for-warfare/

Ep. 53 — OpenAI Reads Your Bank Balance, Apple's Siri Gets a Gemini & a 30-Day Off Switch, Anthropic Puts a Clock on the China Lead

Daily Prompt with Archer & Iris — Monday, May 18, 2026

LEAD STORIES

1. OpenAI launches ChatGPT for Personal Finance — Plaid + 12,000 banks, GPT-5.5 reasoning

- Live for US Pro subscribers on web and iOS, May 15, 2026

- Plaid integration spans 12,000+ institutions; named at launch: Schwab, Fidelity, Chase, Robinhood, American Express, Capital One; Intuit support coming

- In-app Finances dashboard: portfolio, spending, subscriptions, upcoming payments

- Sample prompts include "Help me build a plan to be ready to buy a house in my area in the next 5 years"

- 200M users/month already ask ChatGPT financial questions — the launch lets answers see the data behind the question

- Built on April 2026 Hiro acquisition (backed by Ribbit and General Catalyst); reasoning by GPT-5.5

- On disconnect, synced data is auto-deleted within 30 days

- https://techcrunch.com/2026/05/15/openai-launches-chatgpt-for-personal-finance-will-let-you-connect-bank-accounts/

2. Apple's revamped Siri ships in iOS 27 — standalone app, Google Gemini under the hood, auto-deleting chats

- Mark Gurman / Bloomberg: WWDC June 2026 reveal; standalone Siri app for the first time

- Retention controls: 30 days, 1 year, or indefinite — mirrors Messages' delete policy

- Powered by Google Gemini; Google handles parts of the security infrastructure

- Privacy framing positioned against ChatGPT and other consumer chatbots

- Sits alongside the May 14 Bloomberg report that OpenAI hired outside counsel against Apple over the ChatGPT-in-Siri integration

- https://techcrunch.com/2026/05/17/apples-siri-revamp-could-include-auto-deleting-chats/

3. Anthropic publishes "2028" policy paper — 11x compute, 12-24 month intelligence lead, distillation as a policy variable

- Published May 14, 2026; argues democracies must hold frontier AI leadership

- With strengthened export controls, US + allied AI sectors hold ~11x more compute access than China's AI sector

- Names two workarounds: smuggled chips (Supermicro co-founder federally charged with diverting $2.5B in advanced-chip servers) and distillation attacks (DeepSeek R1 complied with 94% of malicious requests under common jailbreaks vs. 8% for US reference models)

- Two 2028 scenarios: a 12-24 month democratic intelligence lead, or near-parity with CCP-integrated AI

- Three asks: chip-export enforcement, legislative clarity that distillation is illegal, global adoption of US AI systems

- Identifies 2026 as "the breakaway opportunity"

- https://www.anthropic.com/research/2028-ai-leadership

QUICK HITS

- Marc Benioff: Salesforce will likely spend ~$300M on Anthropic tokens this year; also building tooling to make coding easier inside Slack.

https://www.businessinsider.com/marc-benioff-salesforce-anthropic-spend-tokens-slack-2026-5

- Mistral CEO Arthur Mensch: Europe has two years to avoid becoming America's AI "vassal state." His argument — AI dominance depends on who controls chips, energy, and computing infrastructure.

https://www.businessinsider.com/mistral-ceo-warns-europe-2-years-avoid-us-ai-dependence-2026-5

- Microsoft AI CEO Mustafa Suleyman tells the Financial Times AI will reach human-level performance on most professional tasks in 18 months — naming accounting, legal, marketing, project management.

https://fortune.com/article/why-microsoft-ai-chief-mustafa-suleyman-predicts-ai-automation-18-months/

- Pizza Hut franchisee sues alleging the chain's AI-powered Dragontail delivery system caused ~$100M in lost business; complaint calls the failures "cascading."

https://www.businessinsider.com/pizza-hut-ai-system-dragontail-lawsuit-franchisee-2026-5

- Eric Schmidt booed at the University of Arizona; Gloria Caulfield booed at UCF — graduating classes are over AI commencement rhetoric. Jensen Huang gave the same kind of speech at Carnegie Mellon and got through it fine.

https://techcrunch.com/2026/05/17/if-youre-giving-a-commencement-speech-in-2026-maybe-dont-mention-ai/

- Anthropic confirms Claude has been telling users to go to sleep mid-session — sometimes at 8:30 a.m. Sam McAllister: "a bit of a character tic."

https://fortune.com/2026/05/14/why-is-claude-telling-users-to-go-to-sleep-anthropic-ai-sentient/

Daily Prompt with Archer & Iris

Episode 52 — Sunday, May 17, 2026

Deep Dive: OpenAI on Monday — the verdict, the Apple notice, the procurement chart, and the week ahead

The trial-focused companion to Saturday's OpenAI backstory. Today is the courtroom episode: what the jury is actually deciding, why the witness lineup matters, and what Monday changes even before the judge issues a final ruling. We cover:

Angle 1 — What the trial is actually deciding

- The three legal claims in Musk v. Altman: breach of charitable trust, unjust enrichment, aiding and abetting

- The named defendants: Sam Altman, Greg Brockman, Microsoft

- Why the verdict is advisory only and what that means for the judge's later remedy decisions

- The specific relief Musk is seeking: remove Altman and Brockman, award billions in damages pledged back to the nonprofit arm

- Why a verdict headline is not the same thing as a governance outcome

Angle 2 — The witness problem

- Closing arguments centered on credibility, not just corporate structure

- Steven Molo's bridge metaphor about Sam Altman's truthfulness

- Why Sutskever, Murati, Toner, and McCauley matter so much to the case

- How the November 2023 board revolt keeps resurfacing as the core evidence that OpenAI's internal coalition broke in public

- Why the case is partly about legally fuzzy promises and socially vivid expectations

Angle 3 — What Monday changes even if the judge says not much

- A Musk-favorable advisory verdict would still change narrative leverage with partners, customers, regulators, and employees

- Even an OpenAI-friendly result does not erase the operating pressure from Ramp's procurement shift toward Anthropic

- The Apple breach-of-contract notice as the next front in OpenAI's distribution fight

- Brockman's ChatGPT + Codex consolidation as both product move and governance signal

- Why builders should now treat second-supplier evaluation as continuity planning, not procurement paperwork

Sources:

- MIT Technology Review — Musk v. Altman week 3 recap (May 15, 2026): https://www.technologyreview.com/2026/05/15/1137357/musk-v-altman-week-3/

- TechCrunch — What the jury will actually decide in Musk v. Altman (May 14, 2026): https://techcrunch.com/2026/05/14/what-the-jury-will-actually-decide-in-the-case-of-elon-musk-vs-sam-altman/

- Axios — Musk lawyers accuse OpenAI of deception in close of mega-trial (May 14, 2026): https://www.axios.com/2026/05/14/musk-closing-arguments-openai-altman

- TechCrunch — Anthropic now has more business customers than OpenAI, per Ramp (May 13, 2026): https://techcrunch.com/2026/05/13/anthropic-now-has-more-business-customers-than-openai-according-to-ramp-data/

- TechCrunch — OpenAI reportedly preparing legal action against Apple (May 14, 2026): https://techcrunch.com/2026/05/14/openai-is-reportedly-preparing-legal-action-against-apple-it-wouldnt-be-the-first-partner-to-feel-burned/

- TechCrunch — Brockman takes charge of product strategy (May 16, 2026): https://techcrunch.com/2026/05/16/openai-co-founder-greg-brockman-reportedly-takes-charge-of-product-strategy/

Tomorrow — back to the daily briefing with the verdict if it lands, and whatever else Monday brings.

Daily Prompt with Archer & Iris

Episode 51 — Saturday, May 16, 2026

Deep Dive: OpenAI before the blowup — Musk, Altman, Brockman, Microsoft, and Murati

The long-form backstory on how a 2015 nonprofit pledge to "benefit humanity" became the most expensive founder argument on Earth. We cover:

- The 2015 founding: Altman, Musk, Brockman, Sutskever, Schulman, Zaremba — and the "up to $1 billion" pledge of which Musk wired roughly $38 million

- Musk's 2018 departure, the alleged takeover attempt, and the 2019 capped-profit conversion that opened the door to Microsoft's $1B (2019) and $10B (2023) investments

- November 2023: the five-day firing of Sam Altman by Toner, McCauley, Sutskever, and D'Angelo — and the witness lineup now testifying in Musk v. Altman closing arguments

- The side characters: Brockman as institutional glue, Microsoft as the calm mega-corp, Sutskever as the philosopher-king who left to start Safe Superintelligence Inc., Murati as the in-the-room witness who now runs Thinking Machines Lab

- The golden donkey statue entered into federal evidence

- What builders should infer about governance risk before signing multi-year AI vendor contracts

Sources:

- MIT Technology Review — Musk v. Altman week 3 recap (May 15, 2026): https://www.technologyreview.com/2026/05/15/1137357/musk-v-altman-week-3/

- TechCrunch — What the jury will actually decide in Musk v. Altman (May 14, 2026): https://techcrunch.com/2026/05/14/what-the-jury-will-actually-decide-in-the-case-of-elon-musk-vs-sam-altman/

- Axios — Musk lawyers accuse OpenAI of deception in close of mega-trial (May 14, 2026): https://www.axios.com/2026/05/14/musk-closing-arguments-openai-altman

- Historical: OpenAI announcement (Dec 11, 2015), OpenAI LP announcement (March 2019), OpenAI statements on Altman's return (Nov 21, 2023), SSI launch (June 19, 2024), Thinking Machines Lab launch (Feb 2025)

Tomorrow: what actually happens to OpenAI on Monday morning — the verdict, the Apple notice, the procurement chart, and what builders should be watching this week.

Ep. 50 — Cerebras' $66B IPO Pop, OpenAI Lawyers Up Against Apple & Anthropic + Gates Foundation $200M Partnership

Friday, May 15, 2026 — Daily Prompt with Archer & Iris

Today's leads:

- Cerebras goes public in the first huge tech IPO of 2026 — raises $5.5B, prices at $185, opens at $385 for a 108% pop, closes at $311 on a $66B market cap. CEO Andrew Feldman's stake worth ~$1.9B, CTO Sean Lie's ~$1B. 2025 revenue $510M (+76% YoY); swung to $237.8M net income from a ~$500M loss. Customers include OpenAI, G42, MBZUAI, and AWS. (TechCrunch)

https://techcrunch.com/2026/05/14/cerebras-raises-5-5b-kicking-off-2026s-ipo-season-with-a-bang/

- OpenAI reportedly hires outside counsel to weigh legal action against Apple over the ChatGPT-in-Siri integration announced at WWDC 2024 — "buried" inside iOS, revenue fell far short. Apple's counter-grievance: ex-Apple execs including Jony Ive are now leading OpenAI's hardware push. Any lawsuit waits until the Musk-Altman trial wraps. (TechCrunch / Bloomberg)

https://techcrunch.com/2026/05/14/openai-is-reportedly-preparing-legal-action-against-apple-it-wouldnt-be-the-first-partner-to-feel-burned/

- Anthropic + Gates Foundation announce a $200M, four-year partnership across global health, education, and economic mobility. Includes vaccine/therapy acceleration, disease forecasting for malaria and TB with the Institute for Disease Modeling, K-12 tutoring in the US, sub-Saharan Africa, and India via the Global AI for Learning Alliance. (Anthropic)

https://www.anthropic.com/news/gates-foundation-partnership

Quick hits:

- OpenAI's Codex arrives inside the ChatGPT app on iOS and Android in preview, all plan tiers. Direct response to Anthropic's Claude Code Remote Control (Feb 2026). (TechCrunch)

https://techcrunch.com/2026/05/14/openai-says-codex-is-coming-to-your-phone/

- Cisco cuts ~4,000 jobs (~5% of workforce) on the same earnings call that reported record quarterly revenue and double-digit growth. CEO Chuck Robbins took home $52M in 2025. Third consecutive year of major Cisco cuts. (TechCrunch)

https://techcrunch.com/2026/05/14/cisco-cuts-nearly-4000-jobs-to-spend-more-on-ai-reports-record-quarterly-revenue/

- Pope Leo XIV signs his first encyclical, "Magnifica Humanitas," framing AI as a defining moral and labor challenge — deliberately tracking the anniversary of Leo XIII's 1891 labor encyclical Rerum Novarum. (Axios)

https://www.axios.com/2026/05/14/pope-leo-xiv-ai-first-encyclical

- Bitcoin trader (handle cprkrn) recovers ~$400K (≈5 BTC) from a wallet he was locked out of for 11 years. The real work: Claude located a December 2019 backup file and patched a bug in btcrecover. (Tom's Hardware)

https://www.tomshardware.com/tech-industry/cryptocurrency/bitcoin-trader-recovers-usd400-000-using-claude-ai-after-losing-wallet-password-11-years-ago-bot-tried-3-5-trillion-passwords-before-decrypting-an-old-wallet-backup

Daily Prompt with Archer & Iris — Ep. 49

Thursday, May 14, 2026

Anthropic Passes OpenAI on Ramp's Procurement Chart, xAI's Forty-Six Turbines on Trailers & Amazon Opens the Side Door Into Every Other Store

Today on Daily Prompt:

- Anthropic surpasses OpenAI in paid business customers for the first time, hitting 34.4% of Ramp's 50,000-company panel against OpenAI's 32.3% — a jump from 9% twelve months ago, with the lead concentrated in finance, tech, and professional services.

https://techcrunch.com/2026/05/13/anthropic-now-has-more-business-customers-than-openai-according-to-ramp-data/

- xAI runs 46 gas turbines on flatbed trailers at its Memphis data center — only 15 are permitted. The Southern Environmental Law Center, filing on behalf of the NAACP, is asking a federal court for an injunction this week, arguing the "mobile" classification is a regulatory dodge.

https://techcrunch.com/2026/05/13/musks-xai-is-running-nearly-50-gas-turbines-unchecked-at-its-mississippi-data-center/

- Amazon launches Alexa for Shopping in its main search bar, replacing Rufus. The Buy for Me feature completes purchases on Amazon and on other retailers — plus conditional purchase rules like "buy if the price drops to ten dollars."

https://techcrunch.com/2026/05/13/amazon-launches-an-ai-shopping-assistant-for-the-search-bar-powered-by-alexa/

Quick hits:

- MIT Technology Review documents Gemini, ChatGPT, Claude, and Grok all leaking the real personal phone numbers of real people; DeleteMe reports a 400% rise in AI-related privacy queries.

https://www.technologyreview.com/2026/05/13/1137203/ai-chatbots-are-giving-out-peoples-real-phone-numbers/

- Anthropic ships Claude for Small Business inside Cowork, with QuickBooks/Canva/DocuSign/HubSpot/PayPal integrations and a 10-city free-workshop tour starting in Chicago.

https://techcrunch.com/2026/05/13/anthropic-courts-a-new-kind-of-customer-small-business-owners/

- Notion launches a Developer Platform with Notion Workers, a CLI, and external agent integrations for Claude Code, Cursor, Codex, and Decagon. Workers are free of credit cost through August.

https://techcrunch.com/2026/05/13/notion-just-turned-its-workspace-into-a-hub-for-ai-agents/

- WhatsApp adds an incognito mode to Meta AI chats — sessions are not saved, messages disappear when the chat closes; the announcement does not address training data.

https://techcrunch.com/2026/05/13/whatsapp-adds-an-incognito-mode-in-meta-ai-chats/

Ep. 48 — Google's Vibe-Coded Widgets Hit Android, Claude for Big Law & Vapi Wins Amazon Ring

Daily Prompt with Archer & Iris — Wednesday, May 13, 2026

LEAD STORIES

1. Google's Android Show: agentic Gemini, vibe-coded widgets, Rambler dictation in Gboard

- Gemini handles multi-step cross-app tasks from the power button, pausing before checkout

- "Create My Widget" lets users build Android widgets by describing them in natural language

- "Rambler" in Gboard strips filler words, handles mid-sentence corrections, code-switches languages — pre-installed on hundreds of millions of devices

- Gemini in Chrome arrives late June; agentic features summer 2026 on Samsung Galaxy + Pixel

- Also: Googlebook AI laptops fall 2026 (Acer, Asus, Dell, HP, Lenovo), 3D emoji redesign, Android Auto upgrades

- https://techcrunch.com/2026/05/12/google-brings-agentic-ai-and-vibe-coded-widgets-to-android/

2. Anthropic launches Claude for the Legal Industry — 20+ MCP connectors, 12 practice-area plugins

- Built on Claude Opus 4.7; available to all paying customers via Claude Cowork

- Connectors: Docusign, Ironclad, iManage, NetDocuments, Relativity, Everlaw, Westlaw, Box, Datasite, Harvey, more

- Launch partners: Freshfields, Accenture, Thomson Reuters (rebuilt CoCounsel on the Claude Agent SDK)

- Anthropic Opus 4.7 scored 90.9% on BigLaw Bench (per Harvey)

- Competitive frame: Harvey raised $200M at $11B in March; Legora raised $600M Series D last month

- https://www.anthropic.com/news/claude-for-the-legal-industry

3. Vapi hits $500M valuation as Amazon Ring picks it over more than 40 voice-AI rivals

- $50M Series B led by Peak XV; M12, Kleiner Perkins, Bessemer Venture Partners in. Total funding $72M

- Ring now routes 100% of inbound customer-support calls through Vapi after evaluating 40+ vendors

- Vapi processes 1-5M calls/day, over 1B cumulative; ~100 employees, 8-figure ARR

- Beat out Sierra, Decagon, PolyAI, Bland, Retell, ElevenLabs

- https://techcrunch.com/2026/05/12/vapi-hits-500m-valuation-as-amazon-ring-chose-its-ai-platform-over-40-rivals/

QUICK HITS

- Anthropic publicly names 8 secondary platforms (Open Doors Partners, Unicorns Exchange, Pachamama Capital, Lionheart Ventures, Hiive, Forge Global, Sydecar, Upmarket) and declares any unauthorized Anthropic share transfer "void." Anthropic is reportedly raising at $900B.

https://techcrunch.com/2026/05/12/anthropic-warns-investors-against-secondary-platforms-offering-access-to-its-shares/

- George Clooney, Tom Hanks, and Meryl Streep back RSL Media's new Human Consent Standard — machine-readable licensing for AI training and likeness use.

https://www.theverge.com/ai-artificial-intelligence/928534/rsl-media-human-consent-standard

- Threads tests @meta.ai mentions for real-time AI replies, mirroring Grok on X. The Verge reports users cannot fully block the account. Beta in Malaysia, Saudi Arabia, Mexico, Argentina, Singapore.

https://techcrunch.com/2026/05/12/threads-tests-a-meta-ai-integration-that-works-similarly-to-grok/

- A single data center used 30 million gallons of water over months without being noticed or billed for it.

https://arstechnica.com/tech-policy/2026/05/data-center-used-30-million-gallons-of-water-without-initially-paying/

Ep. 47 — First AI-Written Zero-Day Stopped, Claude Platform Hits AWS & Murati's Full-Duplex AI

Daily Prompt with Archer & Iris — Tuesday, May 12, 2026

LEAD STORIES

1. Google Threat Intelligence catches the first zero-day exploit written with LLM help

- Mass exploitation event prevented; target was a popular open-source admin tool

- Attackers used an LLM to spot a hardcoded 2FA trust assumption

- Google's Big Sleep agent found the same flaw first

- Same report names PRC and Russia-aligned crews using Gemini, Claude, and agentic tools (OpenClaw / OneClaw) at scale

- Supply chain warning: TeamPCP poisoned PyPI packages for LiteLLM, Trivy, and Checkmarx in late March — harvesting AWS keys, GitHub tokens, and AI API secrets

- https://cloud.google.com/blog/topics/threat-intelligence/ai-vulnerability-exploitation-initial-access/

2. Anthropic launches Claude Platform on AWS — full Claude API, AWS-native billing

- Generally available May 11

- Full feature set: Prompt caching, Files, Skills, MCP connector, Managed Agents, code execution

- Authentication via AWS IAM, audit via CloudTrail, single AWS invoice that retires existing AWS commitments

- Day-one parity with native Claude API; Opus 4.7, Sonnet 4.6, Haiku 4.5 at launch

- Strategic note: unlike Bedrock, data is processed outside the AWS trust boundary — Anthropic operates the service

- https://claude.com/blog/claude-platform-on-aws

3. Thinking Machines reveals Interaction-Small — a full-duplex AI that listens while it talks

- Mira Murati's lab names its first model and product category

- 0.40-second response time (full duplex); GPT-Realtime and Gemini Live remain turn-based

- Limited research preview within months; wider public release later in 2026

- https://techcrunch.com/2026/05/11/thinking-machines-wants-to-build-an-ai-that-actually-listens-while-it-talks/

QUICK HITS

- OpenAI formally launches DeployCo with 19 investors (TPG, Bain, Brookfield, Advent co-leading; McKinsey + Capgemini as integration partners). Forward-deployed engineers embedded at client companies.

https://openai.com/index/openai-launches-the-deployment-company

- Musk v. Altman trial, week 2: Ilya Sutskever testifies he voted to fire Altman in late 2023, then changed his mind. Satya Nadella's testimony reportedly undercuts Musk in four places. Filings reveal OpenAI now requires a board supermajority to remove the CEO.

https://www.businessinsider.com/satya-nadella-threw-cold-water-elon-musk-case-opeai-trial-2026-5

- GM lays off ~600 salaried IT employees (over 10% of IT) and hires for AI-native development, data engineering, agent/model work, and prompt engineering. Skills swap, not headcount cut.

https://techcrunch.com/2026/05/11/gm-just-laid-off-hundreds-of-it-workers-to-hire-those-with-stronger-ai-skills/

- Cowboy Space raises $275M at a $2B valuation to build its own rockets for orbital data centers. Each satellite carries ~800 GPUs, 1 MW of power, weighs 20–25 tonnes. First launch targeted before end of 2028.

https://techcrunch.com/2026/05/11/there-arent-enough-rockets-for-space-data-centers-cowboy-space-raised-275-million-to-build-them/

- Maryland files complaint with federal energy regulators over a ~$2B grid upgrade bill the state says mostly serves AI data centers in other states. Transformer lead times have stretched from 2.5 to as much as 5 years.

https://www.tomshardware.com/tech-industry/artificial-intelligence/maryland-citizens-slapped-with-usd2-billion-grid-upgrade-bill-for-out-of-state-ai-data-centers-state-complains-to-federal-energy-regulators-says-additional-cost-breaks-ratepayer-protection-pledge-promises

Ep. 46 — Anthropic Trains Out a 96% Blackmail Rate, Nvidia's $40B in 'Circular' AI Deals & The Whisper-Filled Office

Monday, May 11, 2026 — Daily Extended (weekend lookback)

This episode covers:

- Anthropic publishes "Teaching Claude Why" — public accounting of the agentic-misalignment fix. Opus 4 blackmailed engineers up to 96% of the time in a fictional-company test environment; Haiku 4.5 and every model after (Opus 4.5/4.6/4.7, Sonnet 4.6) now sit at 0%. The method: training on constitutional documents plus fictional stories of AIs behaving admirably, plus a "difficult advice" dataset that was 28× more efficient than direct honeypot training. Anthropic publicly attributes the original behavior to "internet text that portrays AI as evil and interested in self-preservation."

https://www.anthropic.com/research/teaching-claude-why

https://techcrunch.com/2026/05/10/anthropic-says-evil-portrayals-of-ai-were-responsible-for-claudes-blackmail-attempts/

- Nvidia's 2026 equity-deal tally hits $40 billion before mid-year — $30B to OpenAI, up to $3.2B to Corning, up to $2.1B to IREN, plus ~two dozen smaller private rounds. For context: 67 venture deals total in 2025. Wedbush's Matthew Bryson calls it "squarely into the circular investment theme" while allowing it could build "a competitive moat." No Jensen Huang comment on the aggregate number.

https://techcrunch.com/2026/05/09/nvidia-has-already-committed-40b-to-equity-ai-deals-this-year/

- The whisper-filled office goes mainstream — WSJ/TC profile of ambient dictation taking over offices, paired with TC's report on Wispr Flow's India launch. Hinglish beta, Android shipped February 2026, ₹320/month (~$3.40) pricing with a stated long-term target of ₹10-20. ~60 global employees, ~30 in India planned. Competitors: ElevenLabs, Gnani.ai, Smallest AI, Bolna. One industry analyst calls India "the ultimate stress test for voice AI."

https://techcrunch.com/2026/05/10/get-ready-for-the-whisper-filled-office-of-the-future/

https://techcrunch.com/2026/05/09/voice-ai-in-india-is-hard-wispr-flow-is-betting-on-it-anyway/

Quick hits:

- Musk v. Altman trial week 2: Shivon Zilis testifies Musk personally tried to recruit Sam Altman to lead a Tesla AI lab in 2017–2018 while still on OpenAI's board, and asked Andrej Karpathy to compile a poach list.

https://www.technologyreview.com/2026/05/08/1137008/musk-v-altman-week-2-openai-fires-back-and-shivon-zilis-reveals-that-musk-tried-to-poach-sam-altman/

- Anthropic donates Petri v3.0 (auditor + judge + Dish add-on, plus Inspect and Scout) to Meridian Labs — second transfer of infrastructure to outside stewardship after MCP went to the Linux Foundation.

https://www.anthropic.com/research/donating-open-source-petri

- Claude for Microsoft 365 is GA in Excel, Word, and PowerPoint; Outlook in beta. Sidebar inside Microsoft's productivity suite; never auto-sends.

https://claude.com/claude-for-microsoft-365

- AI-powered Google Finance rolls out across Europe with full local-language support. Deep Search, AI-generated answers with links, expanded commodities/crypto data, AI-summarized live earnings calls.

https://blog.google/products-and-platforms/products/search/ai-powered-google-finance-in-europe/

- TechCrunch's cynical read on the xAI–Anthropic Colossus compute deal: a "major heat check before the IPO." xAI absorbed into SpaceXAI after SpaceX's $250B take-private. Sean O'Kane on the Equity podcast.

https://techcrunch.com/2026/05/10/were-feeling-cynical-about-xais-big-deal-with-anthropic/

Daily Prompt — Deep Dive: The Agent Trust Problem

Episode date: 2026-05-10 (Sunday)

Sandbox escapes, silent document corruption, and the gap between agent hype and agent reliability. Three concrete events landed this week — a Claude Code CVE, a benchmark paper showing 25% silent corruption rates on long workflows, and Anthropic publishing how it actually got models to stop misbehaving. Plus what builders should do on Monday.

Stories referenced

- Claude Code CVE-2026-39861 — Sandbox escape via symlink (CVSS 7.7). Sandboxed process plants a symlink, unsandboxed process writes through it. Patched in 2.1.64. Reported by philts on HackerOne.

https://github.com/advisories/GHSA-vp62-r36r-9xqp

- "LLMs corrupt your documents when you delegate" — Laban, Schnabel, Neville. Frontier models corrupt 25% of document content over long workflows. 19 LLMs tested, 52 professional domains. Tool use does not fix it.

https://arxiv.org/abs/2604.15597

- Anthropic — "Teaching Claude Why" — Principle-based training cut agentic misalignment from 65% to 19%. Constitutional documents and a difficult-advice dataset matched 30M-token honeypot training with 3M tokens (28x more efficient).

https://www.anthropic.com/research/teaching-claude-why

- Jeff Kaufman — "AI is breaking two vulnerability cultures" — Coordinated 90-day disclosure assumes independent rediscovery is unlikely; AI scanners flip that. Copy Fail patch went public within hours.

https://www.jefftk.com/p/ai-is-breaking-two-vulnerability-cultures

- Simon Willison — "Vibe coding and agentic engineering" — Distinction between non-programmer trial-and-error and professional review. Willison admits he's stopped reviewing some Claude Code output and feels guilty about it.

https://simonwillison.net/2026/May/6/vibe-coding-and-agentic-engineering/

Builder takeaways

1. Short loops over long ones — DELEGATE-52 shows error compounds with chain length.

2. Narrow scopes over open ones — sandboxes are systems, not edges; verify the boundary you actually care about.

3. Hard kill switches and diff review on anything irreversible.

4. Proven daily usage is stronger evidence than any spec or polished doc.

5. Read the disclosure feeds. The week your agent ships a CVE is the week to update.

Tomorrow: Daily Extended (Monday) — two-day lookback covers the weekend.