AI BRIEFING · 2026-08-12

Taiwan's nuclear agency hit by an autonomous AI hack linked to China

plus 9 more stories

Also: SpaceXAI launches Grok 4.6, a near-frontier model built for long-running agents, and DeepSeek ships V4 Pro as its flagship model leaves preview. Every story links to its primary source.

Curated by Gaurav · August 12, 2026

01

Taiwan's nuclear agency hit by an autonomous AI hack linked to China

Financial Times Tech ↗

Researchers at Israeli cyberdefense firm Dream found a hacking tool assembled from two open-source AI agent frameworks running largely on its own inside Taiwanese government systems over four days in early July. Up to eight agents worked simultaneously, mapping 21 government systems, probing for weaknesses, and switching tactics when blocked. The operation compromised at least 85 government accounts, took more than 2,500 personnel records, and spread into Taiwan's nuclear safety agency and at least seven energy companies. Internal communications used Simplified Chinese, and Dream's chief strategy officer called it an unprecedented end to end autonomous attack on a government target.

02

SpaceXAI launches Grok 4.6, a near-frontier model built for long-running agents

Unite.AI ↗

SpaceXAI, the company formerly known as xAI, released Grok 4.6 on August 12, a new flagship model tuned for long-running agents and more ambitious interactive and visual work. The company says it scores 61 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Sol and trailing only Opus 5 and Fable 5, with standout results on the GDPVal and coding benchmarks. Pricing starts at $2 per million input tokens and $6 per million output tokens, which xAI pitches as about half the price of other frontier models. It shipped the same day in the Cursor code editor, the company's Grok Build tool, and its API.

03

DeepSeek ships V4 Pro as its flagship model leaves preview

Unite.AI ↗

DeepSeek released the production build of its flagship model, DeepSeek V4 Pro, ending a preview that ran nearly four months. The version, designated V4 Pro 0813, appeared on August 12 as the general availability release on OpenRouter's model page, and DeepSeek's own API documentation now lists it behind the deepseek-v4-pro endpoint. Pricing carries over unchanged from the preview at $0.435 per million input tokens on a cache miss, keeping it among the cheapest frontier class models available.

04

Google's new Pixel 11 puts Gemini at the center of the AI phone battle with Apple

CNBC Technology ↗

Google unveiled the Pixel 11, 11 Pro, 11 Pro XL, and 11 Pro Fold at its Made by Google event on August 12, weeks before Apple is expected to roll out a rebuilt Siri powered by Gemini models. The phones run a new Tensor G6 chip with 50 percent more TPU power and ship with 12GB of memory as the baseline, which Google says is what lets Gemini run on the device. The pitch is agentic: the assistant can order groceries or book rides, and Magic Cue surfaces reservations and calendar details without being asked. The Pixel 11 Pro starts at $1,099 and the Pro XL at $1,299.

05

Cisco's AI outlook disappoints investors seeking a bigger payoff

Bloomberg Technology ↗

Cisco projected $7.5 billion in sales tied to the AI data center boom this fiscal year, a figure that disappointed investors given the $9.3 billion in AI related orders it booked over the past year. Fourth quarter revenue rose 18 percent to $17.3 billion, beating estimates, with $4 billion of AI infrastructure orders from hyperscalers in the quarter alone. The company guided fiscal 2027 revenue to between $72.2 billion and $73.4 billion. Shares still fell more than 4 percent after hours as traders had priced in faster AI driven growth.

06

AI coding startup Cognition reportedly in talks to raise at a $40B valuation

TechCrunch AI ↗

Cognition is reportedly in talks for a new funding round at a $40 billion valuation, just months after raising $1 billion at a $26 billion valuation in May. The company had reached a $492 million annualized revenue run rate as of May, with enterprise customers including Mercedes-Benz, NASA, and Goldman Sachs. Its Devin agent is positioned to handle programming work such as legacy software updates and platform migrations rather than replacing developers outright.

07

Lovable raises $400M Series C at a $13.3B valuation for its vibe coding platform

Unite.AI ↗

Stockholm based Lovable raised $400 million in Series C funding at a $13.3 billion valuation, announced on August 12, roughly doubling its valuation since its Series B in December 2025. Menlo Ventures led the round, with the EQT managed Scaleup Europe Fund co-leading. The cap table is deliberately global, adding Balderton Capital and Carmignac from Europe alongside Kaszek Ventures and LTS Growth from Latin America. The raise makes Lovable one of the most valuable companies in the fast growing natural language app building category.

08

UK plans to regulate AI in gene synthesis to block bioweapon misuse

Bloomberg Technology ↗

The UK government is planning to regulate the use of artificial intelligence in gene synthesis, according to people familiar with the discussions. Officials are increasingly concerned that the absence of global guardrails makes it easier to use the technology to design biological weapons. Ministers are weighing ways to stop terrorists, other bad actors, and careless researchers from using AI assisted processes to produce synthetic DNA, including new biological weapons legislation.

09

Twitch streamers can now opt out of training Amazon's AI

The Verge ↗

Twitch confirmed that Amazon has been training generative AI models on platform content for years and added a toggle in account security settings that lets users stop it. The setting covers streams, VODs, clips, stream chat, and pictures and text on a channel, and applies only to future model training. It is not retroactive, so anything already absorbed before August 12 stays in. The toggle defaults to on, meaning creators must actively switch it off, and channel owners rather than individual chatters control whether chat is included.

10

Cargo thefts have turned violent in the pursuit of AI hardware

Wired ↗

Two recent incidents in California show criminal rings escalating to confrontation to steal servers and other gear bound for AI data centers, according to a consultant and former Los Angeles sheriff's detective. In one case a driver rear-ended a security escort vehicle in a hit and run to force the shipment to stop. Cargo theft was a $725 million problem in 2025 according to data analytics firm Verisk, and the first three months of 2026 alone brought 767 incidents worth $132 million. Investigators say some of the stolen equipment feeds smuggling networks moving banned US AI technology to China, Russia, and Iran.

← 2026-08-11 All briefings 2026-08-13 →

Tools in this briefing, reviewed in our directory:

GPT-5.6 Sol → OpenRouter → Grok Build → DeepSeek →

Track the tools behind the news:

→ AI Tool Leaderboard → Browse the directory