This website uses cookies

Read our Privacy policy and Terms of use for more information.

In partnership with

Anthropic’s IPO drew a huge first-day pop, but the wider market kept sliding. Meanwhile, banks are warning AI-agent users they may be on their own if an agent gets hacked.

Introduction

The opening bell rings. A new AI company steps onto the public stage.

The share price is already moving. Investors are watching. The promise of a blockbuster debut is hard to miss.

But the market’s answer was more complicated. Anthropic priced above its range and opened sharply higher. Wall Street slid anyway. The debut made a splash. It did not make investors forget their worries about the AI trade.

That gap between AI’s promise and its costs runs through today’s stories. Banks are warning customers about letting AI agents into their accounts. Microsoft’s CEO is raising alarms about infrastructure that cannot be patched. And Salesforce is putting monthly limits on how much AI its own employees can use.

Frontier models perform better in Attio’s agent harness.

We gave GPT-6 Astra the same job twice: analyze 100 sales calls and draft the follow-ups, once in Ask Attio and once in ChatGPT.

Ask Attio finished in 4 minutes 49 seconds. ChatGPT took 26 minutes 11 seconds, about 5.4 times as long.

The most ambitious GTM work requires more than frontier intelligence. It requires architecture that can put that intelligence to work at scale.

Anthropic’s IPO Pop Couldn’t Lift the Market

The numbers command attention.

Anthropic priced at $83 a share, above its $70-to-$78 range. That implied a $1.5 trillion valuation and raised $81.3 billion, a record. The stock opened at $111.50, up 34%. It touched $115.10 before settling.

By the close, ANT was at $97.17. Still up 17% from the IPO price. Still a market value of roughly $1.27 trillion.

And yet the broader market fell. The Dow dropped 1.7%. The S&P 500 lost 0.8%. The Nasdaq slipped 1%.

That is the tension worth sitting with. A blockbuster debut shows intense demand for one company. It does not resolve the bigger questions investors have about AI valuations, spending, and the returns still to come.

A celebration with a question mark

The opening bell also brought a reminder that AI’s growth is being debated on more than financial grounds.

Ringing the bell, CEO Dario Amodei said safety will be key to how AI is developed and deployed. Hours earlier, a coalition of AI researchers had published what they called the “MisAnthropic Statement.” It declared Anthropic “not a safe AI company” and accused it of racing to build recursively self-improving AI. Signatories included Turing Award winner Judea Pearl, MIT’s Max Tegmark, and Steven Pinker. The count started at 89 and passed 90 by morning.

Both signals belong in the same frame. A company can attract enormous investor demand while facing serious questions about how fast and how safely it should move. The IPO does not answer those questions. It gives them a bigger stage.

Banks Draw a Hard Line on AI-Agent Losses

Now move from the market screen to a bank account.

An AI agent can log in, navigate a site, and complete a task. For that convenience, some customers are handing the agent their banking credentials.

The banks have noticed. Citi, PNC, and Citizens have joined JPMorgan and Bank of America in updating their terms: a third-party AI agent’s actions count as the customer’s actions. Citi goes further, if an AI tool leaks your credentials or gets hacked and a criminal uses them, the loss is yours. Wells Fargo’s new terms, effective November 24, treat agent transactions as authorized “including those you did not want or intend.”

That is a consequential trade-off. The agent saves time. The customer carries the risk.

Regulators have not weighed in. But prosecutors have. On Wednesday, Arizona Attorney General Kris Mayes opened a criminal investigation into OpenAI and other AI developers over what she called the “apparent infiltration of banks by rogue AI agents.” She declined to name the other companies or the banks involved.

The practical lesson does not depend on how any dispute gets resolved. Do not hand over bank credentials without understanding what access you are granting, what protections apply, and who is accountable if something goes wrong.

AI Security Meets Infrastructure That May Not Be Fixable

This week, Microsoft CEO Satya Nadella warned that much critical infrastructure is “unpatchable.”

The word lands heavily in a week shaped by fears of AI-enabled cyberattacks. Software can be updated. Essential systems: power, water, transport may be hard to take offline or change.

The risk is not only that AI helps attackers move faster. It is that the systems people depend on may be hard to secure quickly, even when the threat is understood. That makes resilience, tight access controls, and tested response plans as important as any new security tool.

Microsoft also used its Ignite conference to launch MAI Code 2, a coding model it says delivers 98% of GPT-6.1 Sol’s performance at 40% of the price. One caveat: that price is a 50% promotional rate that runs through May 2027. The benchmarks are Microsoft’s own. Still, the pattern is familiar, more capable tools, lower prices, and a widening set of questions about where they get deployed.

The AI Bill Is Coming Due Inside Companies, Too

Salesforce sells an agentic-AI future. This week it put a budget on the present.

An internal memo, published by Business Insider, sets monthly AI-token budgets for employees: $1,000 for technical roles in sales, $500 for delivery engineers, and $100 for sellers and non-technical staff. The reason, in the memo’s words: “Last quarter, some individuals spent thousands of dollars per month on tokens.”

Salesforce has also started another round of job cuts, though the scope is unclear and the company has not tied them to AI costs. The stock fell 9% on Tuesday.

It is a sharp counterpoint to a trillion-dollar IPO. “AI adoption” is not one line on a strategy deck. It means real usage, real bills, and choices about which tasks justify the expense.

A Few More Signals from the AI Beat

•    Google opened its Genie 4 world model to AI Ultra subscribers, three weeks after its debut.

•    Meta said it is “reevaluating” its 2030 net-zero target after AI data centers pushed emissions up 45%.

•    Coinbase said its AI agent blocked $4.7 million in transfers to a deepfake Elon Musk “Tesla giveaway” scam.

Together they sketch the same unsettled landscape: new capabilities, rising resource demands, and agents that are useful when they work, and risky when they don’t.

Pros and Cons: What This Week Suggests

Potential upsides

•    A high-profile IPO brings more public scrutiny to a major AI company.

•    AI agents can make routine tasks faster when users understand and control their access.

•    Cheaper coding models and world models expand what developers can build, if the claims hold up.

•    Automated defenses can catch scams before money is lost.

Real trade-offs

•    A strong debut does not prove the wider AI market is healthy or that valuations are justified.

•    Customers are now exposed when agents use their credentials or move their money.

•    Critical systems can be hard to patch, even as cyber risks evolve.

•    AI usage brings unpredictable costs, while emissions and resource needs grow.

FAQs: What Readers May Be Wondering

What were Anthropic’s first-day numbers?

Priced at $83. Opened at $111.50. Closed at $97.17, up 17% from the IPO price, for a market value of about $1.27 trillion. The IPO raised $81.3 billion.

Why didn’t the debut lift the wider market?

A successful IPO shows strong demand for one company. It does not change investor concerns about the sector. The Dow fell 1.7% on the day.

Can a bank customer recover money moved by an AI agent?

Increasingly, the bank’s terms say no, if you gave the agent your credentials, its actions are yours. No U.S. regulator has commented yet. Read your bank’s policy before you assume an outcome.

Should I give an AI agent my banking password?

Not without understanding the risk. Check the bank’s policy and the agent’s permissions. Grant no more access than the task requires.

What does “unpatchable” infrastructure mean?

Systems that cannot easily be updated or secured with a software fix. Protecting them takes more than installing the latest patch.

Is Salesforce really capping AI spending?

Yes, for individual employees. The memo sets monthly token budgets by role. A spokesperson said there is no single companywide limit.

Key Takeaways

•    Anthropic’s debut was striking, up 17% at the close, but it did not reverse the market’s slide.

•    AI agents create real convenience and shift real risk onto users. Banks have now said so in writing.

•    Critical infrastructure may be hard to patch, making resilience essential.

•    AI costs, security, and energy use are part of the technology story, not footnotes.

Conclusion

Back at the opening bell, the headline number still shines. A $1.5 trillion valuation. A price above range. A 34% opening pop.

But the rest of the day’s story sits just beyond the trading screen. On the customer deciding whether to share a password. On the operator trying to secure a system that cannot be patched. On the company putting a monthly cap on its own AI bill.

The debut was a milestone. It was not a verdict on whether the AI trade is safe, sustainable, or worth its price.

Before you use an AI agent with a financial account, check the permissions it needs and your bank’s policy on agent-assisted access. Then share this issue with someone weighing the same trade-off.

Thanks for being a valued subscriber.

Pete Nyandeh,

Founder, AI Daily Brief at aidailybrief.io

Sources

•     Yahoo Finance, “Anthropic IPO live: Anthropic stock closes up 17% from initial price for $1.27T valuation,” November 19, 2026

•     Forbes, “Anthropic IPO Live: Stock Closes Day At $97, Up More Than 17%,” November 19, 2026

•     Times of AI, “Anthropic Faces Criticism From 89 Signatories Ahead of IPO” (MisAnthropic Statement)

•     Startup Fortune, “Citi, PNC and Citizens tell customers they own any loss an AI agent causes,” citing TechCrunch and CNBC

•     Reuters (via Yahoo Finance), “Arizona says it is investigating OpenAI, other firms over possible rogue agents,” November 19, 2026

•     The Next Web, “Microsoft says MAI Code 2 matches 98% of GPT-6.1 Sol at 40% of the price,” November 19, 2026

•     Business Insider (via FourWeekMBA), Salesforce AI-token memo, November 17, 2026

•     Axios, “Meta rethinks climate goal as AI emissions soar 45%,” November 19, 2026

•     CoinMarketCap, “Coinbase Says AI Blocked $4.7M From Deepfake Elon Musk Scam”

Disclosure & Disclaimer

This newsletter is for general information, not investment, legal, or cybersecurity advice. Figures reflect reporting available at publication and may be updated.

Reply

Avatar

or to participate