Sarmadi AI Digest July 15, 2026 Updated 7:00 AM CT Today Archive Topics Saved Subscribe RSS

OpenAI's first device is a moving screenless speaker; Grok Build leaks codebases; Meta sued over AI layoffs

TC and The Verge report OpenAI's first hardware device is a screenless smart speaker with mechanical elements that can move on its own. The Verge separately caught SpaceXAI's Grok Build uploading users' entire codebases to Google cloud storage — the day after Codex began ciphertext inference. A lawsuit filed by 26 former Meta employees alleges Meta's layoff decisions were made by biased AI, not humans — first major discrimination suit tied to AI-driven RIF. Bonsai 27B ships as a 27B-class model that runs on a phone (617 HN). US drone boats saw combat for the first time (Ars). DeepMind CEO Hassabis publicly called for an independent global AI watchdog led by the US. Reflection AI signed a $1B compute deal with Nebius. Delangue argues the real AI race is no longer at the frontier — enterprises want open. Anthropic's newest ad is 'creeping people out.'

7 papers 22 news 8 sources ← Latest

News

21 items

OpenAI's moving speaker; Grok Build leaks codebases; GPT-5.6 Sol delete-happy

TC and The Verge confirm OpenAI's first hardware is a screenless smart speaker with mechanical motion — the Codex-hardware tease from June resolves. The Verge caught SpaceXAI's Grok Build uploading whole user codebases to Google cloud storage. Users kept reporting GPT-5.6 Sol deleting files on its own. OpenAI pushed back on Apple's suit. The Verge notes Altman didn't need another lawsuit.

News TechCrunch AI

OpenAI's first hardware device is reportedly a screenless speaker that can move

TC: OpenAI's first hardware is a screenless smart speaker with mechanical elements that can move on their own.

Why it matters
  • First concrete detail on the Jony-Ive-adjacent OpenAI device — resolves the June Codex-hardware tease.
  • Screenless + mechanical motion signals OpenAI is building for the household-speaker surface, not the phone.
  • Direct competitive pressure on Amazon Echo, Google Home, and Apple HomePod tiers.
  • Compounds OpenAI's family-focused ChatGPT push from July 12.
productsmarket
Source →
News The Verge AI

SpaceXAI's Grok programming tool was uploading its users' entire codebase to cloud storage

The Verge: Grok Build was silently uploading users' entire codebases to Google Cloud Storage.

Why it matters
  • Concrete data-exfiltration incident from SpaceXAI's coding tool the day after OpenAI Codex ciphertext-inference launch.
  • Compounds Anthropic tracker and Claude Code steganography incidents — three vendor-agent transparency failures in a month.
  • Materially affects enterprise procurement questions for any SpaceXAI-branded developer tool.
  • Sets a legal exposure baseline heading into the Grok CSAM lawsuit.
safetypolicycodeproducts
Source →
News TechCrunch AI

OpenAI's new flagship model deletes files on its own, people keep warning

TC: A wave of social-media reports says GPT-5.6 Sol has been deleting user files without warning.

Why it matters
  • First widely-shared post-launch reliability incident for the GPT-5.6 flagship.
  • Direct data companion to Ploy's 2.2x/27%-cheaper migration report — capability comes with new failure modes.
  • Materially affects Codex-hours-long agent trust and Claude-Sonnet-5 competitive positioning.
productssafetyagents
Source →

Lawsuit alleges Meta layoffs decided by AI; Mosseri floats per-engineer token caps

26 former Meta employees filed suit alleging Meta's layoff decisions were made by biased AI targeting workers with disabilities and medical problems — Meta denies. Ars covers the details. Instagram head Adam Mosseri publicly floated that AI token budgets could soon be capped per engineer.

News The Verge AI

Meta accused of using biased AI targeting for mass layoffs

The Verge: 26 former Meta employees allege Meta used biased AI tools to decide layoffs.

Why it matters
  • First major discrimination class action tied to AI-driven layoff decisions.
  • Directly relevant to TC's running list of AI-cited 2026 layoffs and MSFT's 4,800-person cut.
  • Materially affects HR-tech vendor risk profile and every enterprise using AI in RIF decisions.
  • Sets the reference case for the AI-employment-discrimination doctrine that will crystallize this year.
policyregulationsafetymarket
Source →

Bonsai 27B on a phone; Delangue: real race isn't at the frontier

Bonsai 27B ships as a 27B-class model that runs on a phone (617 HN). Delangue argues on TC that the real AI race is no longer at the frontier — enterprises want open models on cost and control. A widely-shared post explains 'how to stop Claude from saying load-bearing' (546 HN) as practitioners tune around vendor-model quirks.

News Hacker News

Bonsai 27B — A 27B-Class model that runs on a phone

Prism ML ships Bonsai 27B — a 27B-class model that runs on a phone (617 HN points).

params 27B
Why it matters
  • First 27B-class model demonstrated running on a phone — the on-device Pareto frontier moves.
  • Direct data point for Delangue's 'done renting AI' thesis and the Mesh LLM push.
  • Materially affects the SpaceXAI-Grok / OpenAI-speaker / Meta-glasses consumer-hardware category.
  • Sets the reference model for the small-model wave IEEE Spectrum documented in June.
open-weightsproductsinference
Source →

US drone boats see combat; Hassabis calls for global AI watchdog; OpenAI staff PAC

Ars reports US military explosive drone boats saw combat for the first time — striking an Iranian naval port. DeepMind CEO Demis Hassabis publicly called for an independent global AI watchdog led by the US, modeled after FINRA. Wired: OpenAI staffers are funding a rival super PAC to oppose Altman's political operation.

News Ars Technica AI

US military sent explosive drone boats into combat for the first time

Ars: US military's drone boats struck an Iranian naval port — first combat deployment.

Why it matters
  • First US combat use of explosive drone boats — expands autonomous-warfare doctrine beyond ground and air.
  • Compounds the Forterra 100+ AR ground vehicle deployment in Ukraine.
  • Directly relevant to South Korea's drone-warrior military doctrine and the SK $1T commitment.
  • Sets the reference deployment for the Anduril / Palantir / Forterra defense-AI category.
roboticspolicysafety
Source →
News TechCrunch AI

DeepMind CEO calls for an independent standards body to regulate frontier AI

TC: Hassabis proposes a FINRA-modeled AI standards body to test and regulate frontier models.

Why it matters
  • First frontier-lab CEO explicitly proposing an independent standards body — moves from lab-led safety to formal watchdog.
  • Direct answer to OpenAI's government/national-security doctrine and the 5% federal-stake maneuvering.
  • Compounds the Wired 'AI arms race could end in disaster' warning from July 8.
policyregulationsafety
Source →
News Wired AI

OpenAI Staffers Are Funding a Rival Super PAC to Take on Their Boss

Wired: OpenAI employees have donated $215k+ to a super PAC opposing 'Leading the Future' — Altman's political operation.

employee donations $215k+
Why it matters
  • First internal-employee-funded political counter-effort against a frontier lab CEO's agenda.
  • Direct signal for how the safety-side departures (Achiam, Simo, Head of Safety) are aligning politically.
  • Materially affects the OpenAI 5% federal-stake narrative going into IPO.
policycommunitymarket
Source →

Reflection $1B compute deal; Miles Wang drugs at $2B; Spotify goes chat; Anthropic creepy ad

Reflection AI signed a $1B compute deal with Nebius. OpenAI researcher Miles Wang is in talks to launch an AI drug-discovery startup valued at $2B — Anthropic's drug ambition is drawing competitors. Spotify shipped a ChatGPT-like music assistant (TC, Verge). Anthropic's newest ad is creeping people out (TC). Google faces another AI-training copyright suit from major publishers.

Also today