Paintings Subscribe Tiếng Việt

22 November 2026

AI News/This week in AI: agents, money and privacy (22 Nov)

Five things worth your time this week, short and calm: GPT-6.2, LangChain at $3 billion, CrowdStrike buys agent observability, banks push agent risk to customers, and Google keeps chats longer.

Every week brings a flood of AI news. Here are the five things I think are worth a few minutes of your time.

1. OpenAI releases GPT-6.2 Astra and Sol

Astra for demanding work, Sol for everyday tasks, and Astra’s API is about 20% cheaper. Early testers are lukewarm, though. More in OpenAI releases GPT-6.2 Astra and Sol.

2. LangChain raises $300 million at a $3 billion valuation

The Series C was led by CapitalG, Alphabet’s growth fund, at more than double last year’s valuation. The money goes into LangSmith, its platform for running and governing agents, reportedly used by more than 7,000 organisations.

My take: the money is no longer going into making agents, but into running and controlling them. Same trend as story 3.

3. CrowdStrike buys Fiddler AI and Dreadnode

About $495 million (per the WSJ) for two companies: one watches what agents do, the other attacks AI before release. Details in CrowdStrike pays about $495 million to see what AI agents are doing.

4. US banks: let an agent into your account and the risk is yours

Citi, Wells Fargo and others changed their terms: transactions by third-party AI agents are the customer’s responsibility. Analysis in Banks to customers: let an AI agent into your account and the risk is yours.

5. Google keeps Gemini chats about two months, even with history off

According to TechCrunch and others, temporary chats and chats with history turned off are now kept for about two months, up from 72 hours. Google cites safety and security, and says these chats are not used to train models.

Advice: do not paste sensitive data (passwords, customer details, company documents) into any chatbot, even with history off.


The common thread this week: AI agents are leaving the lab and touching real money, real data and real responsibility. The big question is no longer “what can AI do?” but “who controls it, and who is responsible when it goes wrong?”

Sources: The Next Web (GPT-6.2, LangChain), CyberScoop (CrowdStrike), TechCrunch (banks, Gemini).