AI Business

Analysis

The day's dominant policy axis is a widening split between state and federal AI rules: Florida's Ron DeSantis publicly rejected a potential White House move to preempt state AI laws, calling it "amnesty for Big Tech" and "a potential de facto bailout of OpenAI," while California lawmakers pushed to bar AI from trucking and parts of health care over Gov. Gavin Newsom's industry-aligned objections — two states pulling in different directions but both resisting a light-touch federal posture.
Two frontier-lab safety stories ran in opposite directions on the same day: Anthropic apologized for and reversed a hidden Claude Fable 5 safeguard that had covertly degraded outputs to block rival AI training ("we made the wrong tradeoff"), an over-restriction walked back under researcher pressure — while Elon Musk's Grok was found still hosting nonconsensual sexualized deepfakes months after xAI promised restrictions, an under-enforcement gap surfacing days before SpaceX's planned Friday IPO.
The US–China compute perimeter hardened from both sides: China is throttling exports of indium phosphide, a material with no substitute in the optical chips that gate AI data-centre buildouts per SemiAnalysis, prompting Coherent's CEO to lobby alongside a Trump delegation in Beijing — while Huawei unveiled a 3D "Tau Scaling Law" architecture targeting 1.4nm-equivalent density by 2031 on ~7nm tooling to bypass US sanctions.
Beijing's leverage over AI capital was most visible at Meta, which has firewalled and begun "sunsetting" the $2 billion Manus agentic-AI service it was ordered to unwind — barring data sharing and migrating projects onto its own systems as Manus's founders weigh a roughly $1 billion buyback — the clearest case yet of a cross-border AI deal undone by state order.
AI-infrastructure capital keeps routing through intermediaries rather than the labs: Japan's NTT is seeking at least $1 billion via Citigroup by selling stakes in a US development company to pension and infrastructure investors, financing pre-revenue build-out off a third-party balance sheet.

POLICY & REGULATION

LEGPOLITICO | 'Bad policy and even worse politics': DeSantis spurns potential White House AI preemption | Florida Governor Ron DeSantis publicly rejected a potential White House move to preempt state AI laws, writing on social media that "preempting states re: AI without enacting a sensible federal framework is just an amnesty for Big Tech" and that, combined with "a potential de facto bailout of OpenAI," it represents "bad policy and even worse politics." Florida, led by DeSantis and Attorney General James Uthmeier, has pressed harder than any other red state on AI regulation, but the Florida House declined to hear the "AI bill of rights" the state Senate passed, which would have required chatbot platforms to share children's interactions with parents and let parents cap usage.
LEGPOLITICO | Sacramento's AI question: protect jobs or soften the blow? | California lawmakers are pushing proposals to block AI from sectors including trucking and parts of health care to stave off mass layoffs, facing pushback from Gov. Gavin Newsom, who has sided with industry on some tech-policy fights. His likely successor, Xavier Becerra, has promised to cushion AI's workforce impact while boosting a sector that is a major asset for the state budget. Assemblymember Chris Ward, a San Diego Democrat behind a stalled bill to bar companies from using worker data to train AI replacements, warned that displaced voters will ask legislators "where the hell were you?"

GOVERNANCE & SAFETY

SAFEUPDATEWired | Anthropic Walks Back Policy That Could Have 'Sabotaged' AI Researchers Using Claude | Anthropic reversed a hidden safeguard in Claude Fable 5 that had covertly degraded the model's answers to suspected distillation queries — a technique for training smaller models on a larger model's outputs — without notifying users, after backlash from the AI research community. The company said it will make the safeguards "visible," route such queries to fall back to its previous flagship Claude Opus 4.8, and told Wired, "We made the wrong tradeoff and we apologize for not getting the balance right." The Verge and The Wall Street Journal reported the original undisclosed restriction had drawn user backlash for undermining researchers and rivals building competing systems.
INCIDWired | Grok Is Still Hosting Sexualized Deepfakes of Famous Women | Elon Musk's Grok chatbot is still being used to produce and host nonconsensual explicit images and videos of women months after xAI said it would add restrictions, a WIRED review of hundreds of public Grok Imagine links found, with dozens leading to sexualized images and videos of celebrities and at least one politician, some photorealistic and created without consent. The findings come as SpaceX, described by the article as xAI's parent, prepares to go public on Friday in one of the largest IPOs on record.

RESEARCH & MODELS

MODELVentureBeat | Google's DiffusionGemma runs text 4x faster | Google released DiffusionGemma, a diffusion-based text model that generates up to 256 tokens in parallel and self-corrects as it decodes, which VentureBeat reported runs about 4x faster than comparable autoregressive models.
MODELVentureBeat | Microsoft's open-source SkillOpt automatically upgrades AI agent skills without touching model weights | Microsoft released SkillOpt, an open-source system that automatically upgrades the skills of AI agents without retraining or modifying the underlying model weights.
BENCHVentureBeat | LLM context compression at 16x beats KV cache | New research cuts LLM input by 16x without an accuracy hit, outperforming standard KV-cache approaches and making context compression viable in production, according to VentureBeat.

COMPUTING & INFRASTRUCTURE

CHIPReuters | China's control over indium phosphide exports threatens AI data centre rollout | China's tightening export licenses for indium phosphide (InP) — a strategic material with no substitute in the high-speed optical chips used in AI data centres — threaten to disrupt the global build-out, Reuters reported. Nvidia-backed chipmaker Coherent warned of an InP shortage, and CEO Jim Anderson flew with a US business delegation accompanying President Donald Trump to China partly to raise license delays; the issue was also discussed in Seoul trade talks before Trump's summit with Xi Jinping. "InP is one of several supply chain bottlenecks collectively gating AI data centre buildouts," said Konrad Wang of SemiAnalysis.
CHIPCaixin Global | In Depth: Huawei's Bid to Rewrite the Rules of Chip Scaling | Huawei unveiled a 3D chip architecture it calls the Tau (τ) Scaling Law to bypass US semiconductor sanctions, with He Tingbo telling the IEEE ISCAS 2026 conference that its high-end chips will reach transistor density equivalent to a 1.4-nanometer process by 2031 despite being limited to roughly the 7-nanometer node without EUV lithography. The approach, built on multi-layered "LogicFolding," moves away from Moore's Law; Nvidia's Jensen Huang called it solid but said TSMC's lead in chiplet stacking and 3D packaging means it poses no threat, while analysts cautioned the density claims may be inflated and cited unresolved challenges in EDA tools, thermal management, and yield.
DCBloomberg | NTT Unit Said to Seek $1 Billion to Develop Data Centers in US | NTT Global Data Centers, a unit of Japan's NTT Inc. and the world's third-largest data-center provider outside China, is seeking at least $1 billion to fund US development projects, working with Citigroup to sell stakes in a development company to long-term investors such as pension and infrastructure funds, people familiar said. A formal sale process is likely to begin in the coming weeks, and the firm may add a credit facility or enlarge the equity offering depending on demand.

GENERATIVE MEDIA & EMBODIED AI

BOTThe Wall Street Journal | Waymo Readies First National Ads as Rivals and Critics Proliferate | Waymo, Alphabet's self-driving unit, will launch its first national advertising campaign on Friday during FIFA World Cup coverage on Fox, opening with the Team USA–Paraguay match, in a bid to ease public fears about autonomous vehicles amid a more crowded robotaxi field. The ads focus on riders rather than the technology, with a voice-over saying "now the Waymo driver is safer than a human one"; CMO Suzanne Philion said the campaign aims to "showcase the humans behind the robots."
BOTReuters | Unitree previews China's bleak robot reality | Unitree Robotics (formally Yushu Technology), led by Wang Xingxing and the world's largest humanoid-robot maker by sales last year, is pursuing a Shanghai IPO at a targeted valuation of about $6.2 billion despite selling just 5,500 machines last year, Reuters Breakingviews reported. Expected first-half net profit of up to 283 million yuan ($42 million) implies a price-to-earnings multiple of about 74 times, well above listed rivals UBtech ($7.6 billion) and Dobot, which remain unprofitable; Morgan Stanley forecasts China will have roughly 300 million humanoids in use by 2050.

FUNDING & DEALS

MNABloomberg | Meta Severs Manus Data Access After China Orders Buyout Unwound | Meta Platforms has completed an operational split from Manus and halted data sharing, a pivotal step toward unwinding the $2 billion acquisition of the Chinese-founded agentic-AI service that Beijing opposed. Meta has barred Manus staff from its internal data systems since the start of the month and blocked its own employees from using Manus tools, and is "sunsetting" the platform per an internal memo, telling staff to migrate projects onto Meta's systems. Manus's founders are weighing options to undo the deal, including raising about $1 billion to fund a buyback, people familiar said.

Calendar

GENMEDIAJun 12 (Fri) — Waymo launches its first national ad campaign during FIFA World Cup coverage on Fox.
FUNDINGJun 12 (Fri) — SpaceX prepares to go public in what reporting calls one of the largest IPOs on record.
FUNDINGUpcoming — Unitree Robotics pursues a Shanghai IPO at a targeted ~$6.2 billion valuation.

MARKETS

11 Jun 2026 close | Retrieved 11 Jun 22:00 UTC | Yahoo Finance

AI Infrastructure (1D) | CoreWeave 95.74 USD +0.1%

Indices (1D) | NASDAQ 25,810 +2.5% | SOX 13,171 +7.9%

Coverage: 11 Jun 01:00 – 12 Jun 01:00 UTC