AI News

Thursday 13 August 2026

AI news

Manus spins back out from Meta after China blocks its $2B acquisition

AI startup Manus announced on 11 August that it will resume operating as an independent company, roughly eight months after Meta's reported $2 billion acquisition. China's NDRC blocked and ordered the deal unwound in April, citing outbound-investment and technology-export-control violations. Data generated by certain users on or after 29 December 2025 will be deleted 23-24 August (Singapore time); affected users can back up their data first.

Why it matters: A great-power regulator unwinding an already-closed AI acquisition eight months in shows technology-export control now reaches deals well past signing, not just the approval stage.

bloomberg.com

Anthropic signs $9.1B, 20-year compute deal with bitcoin miner Riot Platforms

Anthropic will take 191 megawatts at Riot Platforms' Rockdale, Texas campus under a 20-year, $9.1 billion agreement confirmed 11 August, with optional five-year extensions that could raise total value to roughly $16.1 billion. Riot plans to reach 96 megawatts by December 2027 and the full 191-megawatt buildout by June 2028, turning a former bitcoin-mining site into a two-tenant AI data-centre campus alongside Riot's existing AMD agreement.

Why it matters: A bitcoin miner becoming an AI infrastructure landlord, and Anthropic locking in two decades of Texas grid power while ERCOT tightens scrutiny of new large-scale capacity, shows how far frontier labs will now reach for guaranteed power.

cnbc.com

Corma raises $60M seed to build the first frontier AI lab for defensive cybersecurity

Corma, a San Francisco and Tel Aviv-based startup founded in 2025 by CEO Alon Pluda, raised $60 million in seed funding led by Sequoia Capital, with Khosla Ventures and Coatue also participating, confirmed 10 August. The company is building what it calls the first foundation model purpose-built for defensive cybersecurity, aimed at countering AI-powered cyberattacks.

Why it matters: As AI increasingly appears on the attacking side of cybersecurity, dedicated frontier-lab investment in AI built specifically to defend rather than attack is a distinct and newer category than general-purpose security tooling.

fortune.com

Google unveils Pixel 11 with a 2nm Tensor G6 chip and expanded on-device Gemini

Google's Made by Google 2026 hardware event on 12 August introduced the Pixel 11 lineup and a new Pixel Watch 5. The Pixel 11 runs Google's new Tensor G6 chip, built on TSMC's 2nm process with Gate-All-Around transistor architecture, delivering up to 20% better power efficiency and 3.5x faster local AI execution alongside expanded on-device Gemini Intelligence features.

Why it matters: A new on-device chip generation built specifically for faster local AI execution shows frontier labs' agentic ambitions now extend to what a phone can run without a network round-trip.

9to5google.com

Nonprofit AI

Hong Kong charity's AI chatbot has handled 3,500 migrant-worker exploitation cases

Migrasia, a Hong Kong-based migrant-rights nonprofit, has run roughly 3,500 unique cases through PoBot, its AI chatbot that helps migrant workers identify and act on exploitation, with monthly new cases growing from 198 to 744. Of complaints handled, 29% concern illegal recruitment practices, 10% disputes over employment benefits, and 7% refund fraud; users are concentrated in the Philippines (29%) and Hong Kong (27.5%).

Why it matters: A named nonprofit, a named tool, and a real caseload breakdown is exactly the frontline "who did what, what happened" story this section usually cannot find.

scmp.com

National League of Cities launches AI cybersecurity initiative for under-resourced local governments

The National League of Cities announced on 11 August an initiative with consulting firm CyberAlliance to help under-resourced local governments understand and remediate cybersecurity threats, using Sally AI, a platform that converts technical cybersecurity data into recommendations city leaders can act on and explain to elected officials, insurers, and the public. The launch follows attacks that compromised water utilities in at least a dozen states.

Why it matters: Aimed squarely at organisations with no security staff and no budget, the same capacity gap most nonprofits face when they try to adopt AI safely.

statescoop.com

Beckman Foundation funds a $25K AI Literacy Program for Irvine schools

The Arnold and Mabel Beckman Foundation awarded Irvine Public Schools Foundation a $25,000 Orange County Community Support Grant to develop and launch an AI Literacy Program for educators and students, confirmed 11 August. The district plans to deploy resources to teachers at the start of the 2026-27 school year, with evaluation and refinement running over three years to build a shared AI literacy framework spanning digital literacy, media literacy, and digital citizenship.

Why it matters: A small, fully-specified AI grant from a funder's own announcement is the opposite of the recycled mega-fund stories that keep failing verification in this section.

beckman-foundation.org