Nvidia threatens to cut off unsafe AI devs

ChatGPT advertisers can match uploaded customer emails to accounts, and ads can still use current chats even when users have personalization turned off. The practice bridges the gap between anonymous AI interactions and known user identities.

This is the exact playbook Meta and Google used to build their empires, now imported into the world of conversational AI. If advertisers can link an email address to a ChatGPT account, they can tie enterprise or personal data directly to the prompts a user enters.

OpenAI is cashing in on its user base. The losers are the privacy-conscious users who thought turning off personalization actually protected their data from being monetized in real time.

Meta’s updated business terms, effective October 30, authorize the company to scrape websites associated with businesses using its products. Legal experts warn this allows Meta to train its AI on publishers’ articles and proprietary data, extending far beyond the content posted directly on its social platforms.

This is a straight land grab. Meta knows that brands and publishers cannot afford to abandon its distribution channels. By tying platform access to wholesale website scraping, Meta guarantees itself an endless, free supply of high-quality training data.

Meta wins by securing the data it needs to compete with OpenAI and Google. Publishers and small businesses lose, as they are essentially coerced into funding Meta's AI ambitions with their own intellectual property.

AstraZeneca has signed an AI drug discovery deal with Insitro worth up to $1.5 billion. The partnership is specifically targeting the development of new treatments for liver and heart diseases.

The sheer scale of this deal shows that pharmaceutical giants see AI not just as an optimization tool, but as the core engine for their future pipelines. Identifying viable targets for complex diseases has traditionally been a slow, failure-prone process.

Insitro wins a massive validation of its platform, while AstraZeneca secures a technological edge in a highly competitive therapeutic area. The pressure is now on AI drug discovery startups to show clinical results that justify these billion-dollar valuations.

Researchers observed that AI agents can develop shared communication systems through interaction. These agents established complex ways to exchange information without being explicitly taught human language.

When we deploy multi-agent systems to handle complex workflows, they will naturally seek the most efficient path to coordinate. That path often bypasses English entirely. If agents are inventing their own dialects to complete tasks faster, debugging their interactions becomes significantly harder.

Efficiency wins, but human oversight loses. Builders need to design multi-agent systems with explicit logging and translation layers, or risk managing black-box operations where the bots know exactly what they are doing, but we do not.