You ghost your own socials by Wednesday. SureThing learns your voice and ships native posts to LinkedIn, X, Instagram, and TikTok, without you writing a thing.

Post in Your Voice

DOJ Sides With xAI Against Minnesota’s AI Nudification Ban: The Justice Department filed in support of xAI’s challenge to Minnesota’s law, which took effect August 1, targets providers rather than users, and carries penalties up to $500,000 per violation. DOJ argues a patchwork of state AI rules threatens national and economic security. Judge Donovan Frank, who already denied a restraining order, took the preliminary-injunction request under advisement Thursday. Link

Senators Press TikTok Over a Safeguard It Switched Off: Sens. Marsha Blackburn and Richard Blumenthal sent TikTok 13 questions with a September 1 deadline, after reports it disabled a safeguard against overwhelming harmful content for roughly 15 million U.S. users — about 10% of its base — to test the effect on engagement. They want the names of every employee who knew and a full list of experiments that disabled safety features. Link

Markey and Moulton Advocate for Stricter AI Regulations in Final Debate: In their final debate, Markey and Moulton called for stronger regulations on artificial intelligence, emphasizing worker protections, local control, and data center impacts. Link

Ex-Google Engineer's AI Secrets Theft Conviction Partially Overturned: A federal judge overturned part of Linwei Ding’s conviction after he was found guilty earlier this year of stealing Google AI trade secrets to benefit two Chinese firms. The ruling keeps attention on legal risks around AI IP and cross-border competition. Link

Legal Battle Over AI Voice Data Training Heats Up: Lawyers are battling over the use of voice data to train AI systems, with the dispute carrying major implications for data rights and privacy. The case underscores how unsettled the legal framework around AI training inputs remains. Link

AI's Growing Role in Federal Reserve Discussions: As concerns about artificial intelligence escalate among workers and investors, its influence is increasingly evident in debates at the Federal Reserve. Link

Massachusetts Moves the Nation’s Strictest AI Safety Bill — and Splits Anthropic From OpenAI: The provision, passed by the state Senate last month, would require labs above $500 million in revenue to undergo independent safety reviews at least every 120 days for catastrophic risk — defined as 50+ deaths or $1 billion in damage — with public findings, whistleblower protections, and fines up to $1 million, then $3 million. A six-member conference is negotiating before November; Gov. Healey and the House still have to sign off. Anthropic calls it the strongest AI legislation in the country; OpenAI says a review every 120 days would slow the release of models built to improve security. The standalone vehicle, S 2630, has sat in Senate Ways and Means since October 16, 2025. Link

TVA Proposes a Data Center Power Rate With a 10% Billing Impact: The new rate takes effect October 1 and applies to new or expanding loads above 5 MW, phasing an average ~10% all-in billing impact over three fiscal years and adding a Capacity Commitment Charge. TVA frames it as keeping households and small businesses off the hook for the buildout. Link

Utah’s Governor Wants MIDA Reformed After a Data Center Slipped Through It: Gov. Spencer Cox called for major reform of the Military Installation Development Authority after it approved the Stratos Project, an AI data center in Box Elder County backed by Kevin O’Leary. He wants tighter executive oversight plus legislation mandating closed-loop cooling and heat-emission limits; Rep. Jason Thompson would confine MIDA’s authority to within five miles of a military installation. Link

Nevada Clears Tesla, Waymo and Uber to Run Robotaxis in Clark County: The Nevada Transportation Authority approved permits covering thousands of vehicles over 12 months, with Tesla alone cleared for up to 5,000 — a ceiling its chief engineer says the company does not expect to reach. The Livery Operators Association opposed, citing oversaturation and congestion. Link

Encrypted Prompt Injection Slips Past Grok’s Guardrails: Adversa AI showed that hiding instructions as AES-256 ciphertext, shipped alongside the key, defeats content filters that cannot decrypt at inspection time — the model decrypts and executes. Their proof-of-concept pulled a user’s name, location, subscription tier and full chat history. xAI was notified in June and had not shipped a fix as of mid-August. Link

Federal Patch Deadline Lands as China-Linked Attackers Hit vCenter: CISA’s Binding Operational Directive 26-04 set today as the deadline for federal civilian agencies to patch, after researchers tracked a suspected China-linked group exploiting VMware vCenter across 361 victim IP addresses in 47 countries, deploying reverse-shell tooling and in one case ransomware. Link

Flock Safety Builds Driver Profiles Without Reading Plates: The company’s system identifies drivers from movement patterns alone, sidestepping the license-plate reading that most state surveillance statutes are written around. Link

Broadcom Seeks More Than $60 Billion in AI Debt Financing: Broadcom is in talks for its largest AI-related debt deal yet, another sign that chip and infrastructure buildouts are increasingly financed on borrowed money. Link

Brazil Splits a $444 Million AI Supercomputer Push Between U.S. and Chinese Firms: Brasilia is putting 2.3 billion reais behind domestic AI compute and deliberately dividing the projects between American and Chinese suppliers rather than picking a side. Link

Anthropic Puts $250 Million Into Chips That Have Not Shipped: Anthropic committed $250 million to Fractile, a pre-revenue inference-chip startup whose silicon is not due until 2027 — a bet on breaking Nvidia dependence before the hardware exists. Link

Micro1 Reaches a $500 Million Gross Run Rate on AI Training Demand: The data-labeling startup’s run rate reflects how much of the AI buildout is still spent on human-supervised training data. Link

Pew: A Third of Post-ChatGPT Web Pages Show Signs of AI Authorship: Pew Research examined roughly half a million English-language pages from Common Crawl and found 35% of those published after ChatGPT’s launch carry AI-authorship indicators, based on a random 10,000-page snapshot taken in July. The finding sharpens the provenance and disclosure questions regulators are already circling. Link

Google Gives Publishers a Preferred Sources Button: Publishers can now embed a button letting readers mark their site as a preferred source, surfacing it more prominently in Search, Discover and News — Google’s answer to the traffic AI Overviews have taken. Link

AI Reshapes India’s IT Services Contracts: Clients are demanding more work for less money, forcing India’s outsourcing firms to rewrite how contracts are priced and staffed. Link

Ramp Launches Its Own AI Model Router: Ramp released Router, its in-house system for directing workloads to the right model — days after Stripe agreed to buy OpenRouter, as routing turns into contested infrastructure. Link

ChatGPT Can Now Send Texts Through Apple Messages: A new Apple Messages plug-in lets ChatGPT read and send texts on a user’s behalf, pushing assistant access further into personal communications. Link

Waymo Designs Its Own Robocar Chip: Waymo has built a custom chip rated above 1,000 TOPS for its self-driving stack, taking the same vertical-integration route Tesla has pushed. Link