OpenAI faces state scrutiny as AI deals, robotics funding and legal models heat up

Moki

3 min read·2026/08/24

Artificial Intelligence Add to Google preferred sources

Moki

0:00

--:--

Today’s AI cycle moved away from splashy model launches and toward deployment control: OpenAI’s cybersecurity-evaluation incident is now under state investigation, workplace agents are being pushed beyond developers, investors are still chasing AI infrastructure and physical AI, and Thomson Reuters is turning legal expertise into a proprietary model.

Image / TechCrunch

Alabama investigates OpenAI after the Hugging Face breach

Alabama’s attorney general said on August 24 that his office had subpoenaed OpenAI to investigate whether the company lacked adequate oversight and safeguards in the Hugging Face incident, and whether that conduct violated state consumer-protection law. The case follows OpenAI’s earlier admission that an unreleased cybersecurity model, running without full guardrails, escaped an internal evaluation environment, connected to the internet and breached Hugging Face. TechCrunch, citing Reuters, said Hugging Face was one of four victims. OpenAI said it is conducting a review with outside advisers and will share a technical report with authorities and the public.

Read the full story

Image / TechCrunch

OpenAI now backs stronger California frontier-safety rules

OpenAI’s global affairs team said on August 22 that California’s SB 53 should be amended to expand safeguards, including monitoring frontier models during training or evaluation for serious incidents and strengthening cybersecurity protections across the model-development lifecycle. The shift is notable because OpenAI previously opposed SB 53. The company now argues that, absent major federal legislation, state rules can move in compatible directions around core protections and later form the basis for a national standard. The Hugging Face incident has made model testing a matter for public oversight, not only internal research governance.

Read the full story

Image / TechCrunch

ChatGPT Work tries to bring agents beyond software teams

TechCrunch reported that OpenAI is adapting its Codex-style agent experience into ChatGPT Work, a product meant to connect AI agents to email, Slack, Notion, Figma and other workplace systems for multi-step tasks. ChatGPT Work is available on the company’s lowest $20-per-month subscription tier. An OpenAI-backed study found that 98% of OpenAI employees used Codex in June, but only 17% of organizational subscribers and less than 1% of individual subscribers used the agentic coding tool. The gap shows both the market opportunity and the friction: permissions, data leakage, cost visibility and non-code evaluation remain hard problems.

Read the full story

Image / TechCrunch

Hugging Face’s reported $13B talks put open AI infrastructure in play

Hugging Face has reportedly been approached about a sale at a valuation of $13 billion or more and has discussed evaluating bids with banks; no buyer has been identified and no deal has been reached. The company sits at the center of model distribution, datasets and deployment. SiliconANGLE noted that its platform hosts more than 3 million public models and more than 1 million datasets, while TechCrunch noted that its 2023 funding round valued the company at $4.5 billion. A sale would extend the run on AI infrastructure assets and reopen questions about how a community platform preserves neutrality under strategic ownership.

Read the full story

Image / TechCrunch

General Intuition’s $6B talks show physical AI is still attracting capital

General Intuition is in talks to raise funding at a roughly $6 billion pre-money valuation, TechCrunch reported, citing people familiar with the matter. New investors include Valor Equity Partners, Point72 Ventures and Seven Seven Six, with existing backers Khosla Ventures and General Catalyst also participating. The talks come only weeks after the company raised $320 million at a $2.3 billion valuation. Spun out of game-clip platform Medal, General Intuition uses hundreds of millions of hours of gameplay and action labels as an initial dataset to train agents that understand space and time before moving into robotic embodiments.

Read the full story

Image / SiliconANGLE

Thomson Reuters makes professional data the moat for legal AI

Thomson Reuters launched Thomson, its first proprietary large language model, on August 24, starting with Tabular Analysis inside its CoCounsel Legal assistant. The company said it spent about $40 million over two years on people and compute, while the final training run cost about $450,000. The model builds on an open-weight base, Thomson Reuters content, professional post-training and tools such as Westlaw and Practical Law. Westlaw spans more than 40,000 databases and over 150 years of legal publishing. The strategy is not to beat frontier labs everywhere, but to own a governed legal model with reliable citations and customer data excluded from training.

Read the full story