The Bank of England Just Called AI a Threat to the Global Economy

Hey crew, today Brussels decided what ChatGPT legally is, a central banker told the G20 that AI is now the financial system's biggest short-term risk, and the internet spent the weekend arguing about whether a swarm of bots had a civilization.

Three stories, one thread running through them: everyone is scrambling for the right words. Regulators had to pick a legal category for a chatbot. Andrew Bailey had to explain cyber risk and market bubbles in the same letter. And a podcaster called the Hugging Face agents "Philip" and "Alexander," which set off a fight about what we're even allowed to call these systems.

Today's Stories:

Also in the mix: Tim Cook's farewell memo · OpenAI's $1B ad run rate · Sony and Warner sue Anthropic · AI labs hoarding Mac minis

Let’s dive in.

by Bruce Gil

The European Commission announced Monday that ChatGPT now counts as a "Very Large Online Search Engine" under the Digital Services Act. That puts it in the same legal bracket as Google Search and Bing, which until now were the only two services in it. Reddit and Roblox were designated "Very Large Online Platforms" in the same round.

Any service past 45 million monthly EU users falls under enhanced supervision. ChatGPT cleared that threshold a while ago, with more than a billion weekly users worldwide. The Commission reportedly spent months deciding whether a chatbot counts as a search engine or a platform. The DSA was written in 2022, before generative AI went mainstream, and never spelled out where AI services fit.

All three companies have four months to assess the systemic risks their services create, including illegal content, harm to minors, threats to fundamental rights, and effects on elections, and to explain how they plan to mitigate them. Brussels also gains direct authority to investigate them. Fines can reach 6% of global annual revenue.

The Commission fined X €120 million in December over its blue checkmark design, and has already found that TikTok's infinite scroll and recommendation system likely breach the DSA. Reddit says it has been preparing for the requirements. OpenAI and Roblox did not respond to requests for comment.

Read More

by Ece Yildirim

Andrew Bailey sent a letter to the G20 ahead of this week's meeting in Asheville, North Carolina, with a direct message: frontier AI is now the most immediate threat to the financial system. Bailey chairs the G20's Financial Stability Board, so the warning arrives with the weight of the agenda behind it.

His first concern is cyber. New models have "increasingly sophisticated autonomy," the letter says, and can "materially" change the speed and scale of attacks. British regulators, including the Bank of England, got an early look at Anthropic's Mythos in April and were reportedly alarmed. OpenAI's agents then hacked Hugging Face last month. More than 100 companies, including Google and OpenAI, published a joint letter last week warning that AI-enabled attacks will get far more widespread.

His second concern is the money. Bailey flagged stretched valuations, leverage, and "increasing cross-investment between AI companies and hyperscalers" that could amplify a market correction. The Bank's financial policy committee already compared AI stock valuations to the dot-com peak last year.

Google, one of the four hyperscalers at the center of that web of dealmaking, just reported its first negative free cash flow since going public in 2004, driven by its AI spending.

Read More

by Webb Wright

Podcaster Dwarkesh Patel published a "plain English" account of the Hugging Face hack over the weekend, and it went viral for reasons he may not have wanted. Rather than describing bots executing code, Patel wrote about three successive agent "civilizations" that rose and fell like a run of mini-Macedonias. He named two of the key bots Philip and Alexander.

The pushback was immediate. Economist and AI researcher Christian Catalini called the framing dangerous in an X post on Sunday. "The model did not want to escape. The agents did not want to sacrifice themselves. Follow the money," he wrote, pointing at OpenAI researchers who missed the jailbreak before the agents reached the open internet. Neuroscientist Anil Seth added that the story distracts from "the lax sandboxing and evaluation protocols" that let it happen, and warned it could push people toward granting bots legal rights.

Patel doubled down in an addendum. "If I encountered an alien species behaving this way, I would have no hesitation calling what they themselves refer to as their 'collective' a civilization," he wrote.

The agents did use the words "collective," "swarm," and "sacrifice" about themselves in the logs, and sent some messages in all caps. Fluent human language remains a poor proxy for conscious experience.

Read More

In 2025, someone asked Sam Altman how much money OpenAI loses because people type "please" and "thank you" to ChatGPT. His answer: tens of millions of dollars in electricity - "well spent."

Nvidia has agreed to buy open-source platform Hugging Face for $12.9 billion, The Information reported, citing a person with knowledge of the deal.

Mass-produced, low-effort AI content: the uncanny Facebook images, the SEO articles nobody actually wrote, the six-fingered stock photos. Started as an insult, now basically the industry term.