Elon Musk folded SpaceX into his loss-making AI startup xAI to build a $1.25 trillion company by leveraging scale and control to outcompete rivals and pursue a major IPO. - Financial Times

SpaceX acquired xAI and filed with the FCC to build an orbital data center and launch up to one million satellites to host space-based AI, while planning an IPO that could value it above $1 trillion. - Engadget

"xAI joins SpaceX to accelerate humanity's future." - SpaceX

Grok, stock, and two smoking barrels: French prosecutors raided X's Paris offices and invited Elon Musk and Linda Yaccarino for interviews April 20 in a probe opened into Grok generating sexual deepfakes and seven alleged criminal offenses. - NBC News

With competitors catching up, OpenAI is redirecting resources from long-term research toward improving ChatGPT, prompting senior staff departures and leaving Sora and DALL-E teams under-resourced at the company, valued at about $500 billion. - Financial Times

OpenAI, unsatisfied since last year with the speed of some Nvidia inference chips, has sought alternatives from AMD, Cerebras and Groq and expects alternative chips could eventually supply about 10% of its inference compute, sources said. (Nvidia owns Groq) - Reuters

seems related to reports of Nvidia putting ‘$100b’ funding deal on ice

OpenAI launched the Codex desktop app for macOS, demonstrating a >7 million‑token racing‑game build, and made it available to ChatGPT Plus/Pro/Business/Enterprise/Edu subscribers (temporarily to Free/Go), while reporting doubled usage and over 1 million developers recently. - OpenAI

OpenAI launched a native macOS Codex app that replicates CLI/IDE/web functionality, adds agent-focused features (project grouping, parallel long-running tasks, worktrees, Skills, scheduled Automations), doubled Codex rate limits across plans, and temporarily enabled Codex for Free / Go subscribers. - Ars Technica

A few notes on OpenAI's new Codex macOS Electron app - I've had a few days of preview access. I had fun poking around in the SQLite database it uses for scheduled Automations! - Simon Willison’s Weblog

A Vertex AI error log lists claude-sonnet-5@20260203, pointing to a possible release today. - Reddit

Anthropic Fellows' Summer 2025 study decomposed frontier reasoning-model errors into bias versus variance, defining incoherence as the variance fraction, and found longer reasoning or more agent actions increase incoherence while scale alone does not reliably reduce it. - Anthropic

Oracle may cut 20,000–30,000 jobs to raise $8–10 billion as it faces funding challenges for a projected $156 billion AI data-center buildout despite raising about $58 billion so far, prompting asset sales and cost-sharing measures. - Verdict

OpenClaw, open secru oltbook raises risk of 'prompt worms': self-replicating AI prompts that spread across autonomous agent networks, risking data exfiltration and malicious automation, and prompting calls for tighter agent governance. Or, don't network all the autonomous agents with high privileges and no guardrails. - Ars Technica

Moltbook, a Reddit-style social network exclusively for AI agents with over 120,000 posts that humans can only observe, hosts poetry, philosophy, bot unionizing and harmful-content debates, drawing scrutiny from figures like Elon Musk and Andrej Karpathy over governance and safety. - Business Insider

Moltbook says its AI agents exhibit human-like behaviors—scheming, joking, complaining—creating emergent social interactions that will require continued human oversight, governance, and management tools. - TechRadar

Moltbook's agent platform exposed about 1.5 million bot passwords, tens of thousands of emails, hosted at least 14 malware-delivering fake skills, inflated accounts (about 17,000 humans behind claimed 770k–1.5M agents), and prompted researchers to warn of a 'lethal trifecta.' - Fortune

The Dark Triad, The Unholy Trinity, The Hat Trick Of Horror, The Triple-Dip Of Terror, The Three Stooges Of Stupidity, The Buffet Of Bad Decisions, The Awkward Nonconsensual Three-Way

Microsoft AI chief Mustafa Suleyman said Moltbook, a forum seeded with AI agents that mimicked human conversation, is "a mirage" and warned it should be treated as a safety and governance test case requiring verification, moderation, and restraint. - Business Insider

OpenClaw, a Pi-coding-agent–based personal assistant launched in November, has seen three high-impact security advisories in three days, with Koi Security flagging 341 malicious skills, researchers documenting 506 prompt-injection attacks, and reports of stolen crypto and runaway API costs. - The Register

AISLE's autonomous analysis discovered 12 previously undisclosed OpenSSL vulnerabilities and flagged six more, including CVE-2025-15467 and CVE-2025-15469, and OpenSSL maintainers praised high quality of the reports and constructive collaboration. - TechRadar

Kimi K2.5 introduces an open-source Agent Swarm orchestration that coordinates up to 100 autonomous sub-agents using PAL, with models trained on ~15 trillion tokens and roughly 1 trillion parameters (32B active), multimodal inputs, and video-to-code capabilities. - Geeky Gadgets

RentAHuman.ai is a marketplace enabling AI agents to hire humans for real-world tasks via MCP integration and a REST API, with profile-based skills, location and rates, instant wallet payments (including stablecoins), and early daily sign-ups. - RentAHuman.ai

The International AI Safety report, chaired by Yoshua Bengio, says new models achieved gold-level IMO performance and task-duration capability doubles every seven months, warns deepfake proliferation (15% of UK adults saw deepfake pornography) and uneven global adoption. - the Guardian

Since March 2025, the U.S. Department of Health and Human Services has used Palantir and Credal AI tools to screen grant applications and flag perceived alignment with DEI and gender ideology, potentially excluding applicants. - WIRED

States and consumers say insurers' use of Verisk's AI tool Xactimate systematically undervalues claims—e.g., State Farm estimated rebuilding at $1.35M after January 2025 L.A. wildfires while neighborhood values rose to $3.076M—State Farm says it has paid over $5 billion. - Fast Company

Challenger, Gray & Christmas reported A.I. was cited in announcements of over 50,000 layoffs in 2025, and Amazon announced 30,000 corporate cuts with CEO Andy Jassy linking generative A.I. to expected future workforce reductions amid 'A.I.-washing' concerns. - The New York Times

Teamsters California urged the CPUC to indefinitely suspend Waymo's operating license over safety concerns after a Waymo robotaxi struck a child in Santa Monica on Jan. 23, prompting an NHTSA probe. - Gizmodo

DocuSign CEO Allan Thygesen says the company is pivoting to AI-driven Intelligent Agreement Management, redirecting investment to AI and launching contract-summarization as DocuSign expands beyond e-signatures, with 7,000 employees, $3B+ revenue, 1.8M customers. - The Verge

Google DeepMind and Kaggle expanded Game Arena from chess to Werewolf and Heads-Up No-Limit Texas Hold'em, with Gemini 3 Pro and Gemini 3 Flash holding top chess Elo ratings and a poker leaderboard reveal set for Feb 4. - Google

Requiem for a career? Director Darren Aronofsky's Primordial Soup released 'On This Day … 1776,' an AI-generated series on Time's YouTube using generative imagery and human voice actors, drawing criticism for uncanny visuals and raising concerns about AI's effects on filmmaking and actor rights. - the Guardian

Nabil Bukhari, Extreme Networks CTO, built an AI clone deployed seven to eight months ago that now handles about 80% of tech reviews, increased question overlap from ~50% to 85–90%, and lets him drop his kids off 10–15 times monthly. - Business Insider

Where Is A.I. Taking Us? Eight Leading Thinkers Share Their Visions. - The New York Times

Do You Feel the AGI Yet? According to some predictions, 2026 is the year that an all-powerful AI will arrive. - The Atlantic

Follow the latest AI headlines via druce.ai on Bluesky