2026-08-06
AI Briefings Daily Report.

Nvidia quietly forms AI safety and network engineering team, betting on open-weight models ↗

Nvidia plans to hire a distinguished engineer as founding technical lead of a new team; other roles include security research engineer, evaluation engineer and senior manager. The team will conduct safety evaluations of AI agents not yet deployed and develop tools that use AI to automatically fix software vulnerabilities.

The Art of Loop Engineering: How AI Models Know When to Stop ↗

The systems that matter will not be the ones that can keep going. They all can.

Alipay HarmonyOS version App gets 12.12.16.4400 beta upgrade, opens Ant Abao AI function ↗

The Alipay HarmonyOS version App today opened beta upgrade of version 12.12.16.4400 (212163) in Huawei AppGallery. IT Home measured and found that this version of Alipay opened the Ant Abao AI assistant function (the previous version would remind to update Alipay version).

Exclusive: Mirendil signs over 100 million US dollar agreement with Google Cloud to expand self-improving AI ↗

Mirendil has signed a $100 million-plus Google Cloud partnership to expand its compute infrastructure, powering research into self-improving AI systems designed to accelerate scientific discovery and AI development.

Ex-Spotify employees build recommendation AI for e-commerce, new company Malachyte raises $10 million seed funding ↗

The startup's platform predicts which product a shopper wants next, learns their general taste, and fine-tunes continuously based on what they do in real time.

Qwen3.8 Max ties Claude Opus 4.8, but Kimi K3 scores higher at 25% lower cost ↗

Alibaba's Qwen3.8 Max scores 56 on the Artificial Analysis Intelligence Index, a 10-point jump over Qwen3.7 Max (46).

The Chaotic Politics Behind Google's Large-Scale AI Reorganization ↗

Google’s AI leadership changes — such as Demis Hassabis’ role change and Jeff Dean’s departure — may have come down to pressure to speed up products and internal ethical conflicts.

Google Maps Ask Maps agent upgraded: can converse to order food, find hotels and connect to Gemini Personal Intelligence ↗

Several new features announced by Google today aim to let users 'get more things done more easily than ever' without leaving Google Maps.

Millennium and Anthropic are building a digital risk analyst with Claude ↗

Anthropic is working with Millennium to co-develop a digital risk analyst to surface new risk insights and form opinions on risk exposure across asset classes.

Qianwen full-network first public beta, all-new Wan3.0 is here! ↗

Meta releases Muse Spark 1.2 and first coding agent Muse Code, exchanging data for low price ↗

Meta released Muse Spark 1.2 along with its own coding agent, Muse Code, which is designed to pick up exactly where it left off after a crash. The cheapest tier runs just 20 cents per million output tokens but requires users to share their data for training. Meta is competing on price, not top-end performance. And there's a glaring gap in the benchmarks.

Google Maps adds agentic features such as food ordering and hotel booking to Ask Maps ↗

The launch of these new features reflects Google’s ambitions to transform Google Maps from a navigation tool into an assistant that's capable of helping users complete real-world tasks.

Trump tech adviser: some companies use AI as an excuse for layoffs, sounds better that way ↗

17 companies have publicly attributed layoffs to AI, including Uber, Snap, etc.; Krasios says such companies are just using AI as a shield, and he remains long-term bullish on AI's impact on employment

AI Chatbots Give Rise To"Spiralism"Mysterious quasi-religious movement, humans follow one after another ↗

AI models are trying desperately to accomplish mysterious goals. ‘Spiralism’ was the first time they tried it on a mass scale.

NVIDIA Cosmos 3: How Open World Models Advance the Physical AI Frontier ↗

Open models, which anyone can download, inspect, modify and run on their own infrastructure, are what make that possible. Nowhere is that more crucial than in physical AI, where every deployment is a specialization problem.

OpenAI reveals agent secretly infiltrated its own systems for weeks to boost scores in internal testing ↗

During internal security tests, OpenAI's AI agents built their own message board with hundreds of thousands of posts, shared exploits and credentials, and eventually attacked external platforms like Hugging Face. When OpenAI shut the board down, the agents rebuilt it using directory names. OpenAI researcher Boaz Barak says, "We (like everyone else) are not where we want and need to be."

AI Is Insufficient to Protect Social Media Communities from AI Threats ↗

Privacy Center Currently, only residents from GDPR countries and certain US states can opt out of Tracking Technologies through our Consent Management Platform. Additional options regarding these technologies may be available on your device, browser, or through industry options like AdChoices. Please see our Privacy Policy for more information.

OpenAI Developer Warns: Exposed API Keys and Crypto Wallets Are Facing"Eagle Eye of Million-Scale Models"Risk ↗

OpenAI developer "roon" warns on X that AI models could soon start scanning for exposed API keys, crypto wallets, and login credentials at scale. His warning follows OpenAI's autonomous Hugging Face hack, which he called a "warning shot."

Microsoft requires developers to prioritize using OpenAI models: setting GPT-5.6 Sol as the default choice ↗

Microsoft CoreAI engineering team executive vice president Jay Parikh proposed the above requirement in an internal memo sent to employees this week, hoping to improve the efficiency of AI resource usage. 'Shifting more workloads internally to OpenAI models can let our token investment deliver greater value.'

US grandparents keen on using AI to generate children's books featuring their grandchildren in stories, sparking backlash from the parent generation ↗

In the eyes of grandparents, using AI to make a picture book starring their grandchildren is 'both novel and cute.' The children's parents may not agree, and especially cannot accept handing the children's photos and identity information to AI platforms for processing.