I never enjoyed suspense films, and the last time I watched a horror film was probably more than 25 years ago. Last night, however, I sat down in front of the TV and somehow decided to watch The Housemaid, the psychological thriller with Sydney Sweeney and Amanda Seyfried.
While Netflix’s signature “tudum” chimed, I pulled out my phone for a last scroll through my X feed, as one does these days, where I saw that Anthropic’s chief executive Dario Amodei had just published an essay titled “Policy on the AI Exponential” (link at the end of the newsletter), so I started reading, casually… I did not finish it.
The evidence of AI’s power and its risks has become undeniable in recent months, wrote Mr Amodei; his own restricted Mythos model demonstrated that frontier systems can disrupt the financial sector, critical infrastructure and national security. The first arrivals are cyber risks; biological risks may soon follow, and serious autonomy risks may not be far behind.
He argues that frontier models should be tested and audited like aircraft, with a regulator empowered to ground a release that fails the review. He welcomes last week’s White House executive order while recommending that Anthropic’s own legislative proposal go further; and a time may come, perhaps relatively soon, when the better comparison for the most powerful systems is “weaponizable nuclear materials”.
This is roughly where I reached for the remote… A thriller screenwriter would build a similar escalation structure, but this is the chief executive of a frontier lab asking to be regulated more firmly than the White House proposed, in the same week his company shipped its latest and most powerful AI model, Claude Fable 5, with practically as many restrictions as new features.
Last night, the thriller was the calmer option.
I went back today (in broad daylight and surrounded by people) and finished reading the essay: all five chapters, from public safety to tax policy to the balance of power between state and citizen, so that you do not have to do it tonight, and can instead watch the opening match of the FIFA World Cup with a clear conscience while scrolling through this issue of The Compute.
The fable and the moral
On June 9, Anthropic released Claude Fable 5, the first Mythos-class model available to the public, priced at $10 per million input tokens and $50 per million output: double the previous flagship rate. Fable 5 shares its underlying weights with the Claude Mythos 5 restricted release; the difference is that Fable 5 intercepts queries touching cybersecurity, biology and model distillation, and quietly routes them to the older Claude Opus 4.8. The restrictions extend to data privacy: every Mythos-class request, including on Fable 5, carries a mandatory 30-day data retention period (in contrast with existing zero-data-retention agreements), which makes this a matter of regulatory and legal compliance much more than engineering fine print.
Researchers who combed through the 319-page system card found a provision for silently degrading responses to requests about frontier AI development, with no notification to the user. They coined the phrase “secret sabotage”; within a day Anthropic made the safeguards visible and apologised, telling Wired: “We made the wrong tradeoff and we apologize for not getting the balance right.”
How to tax the wealth AI creates?
Mr Amodei devotes a full chapter to it, suggesting that income support for displaced workers could be financed by taxing the companies and capital gains the technology enriches. He is not alone this week. The Financial Times argued today that existing tax codes are simply not built for an economy where the gains flow to capital and the income-tax base shrinks; Pictet Asset Management called wealth taxes “almost inevitable” given how concentrated AI gains have become.
The options on the table span the spectrum. Mark Cuban wants a federal levy of under 50 cents per million tokens at the provider level, estimating $10bn a year to start. Senator Bernie Sanders proposes that frontier labs transfer half their equity to a public sovereign wealth fund. And President Donald Trump confirmed on June 5 that his administration has discussed voluntary government stakes in AI companies, an idea Sam Altman pitched himself.
None of these proposals will pass in its current form; together they tell us that the fiscal treatment of AI is now in play, on both sides of the equation, just as the sector heads to public markets.
Brussels takes with one hand and gives with the other
For our European readers, the policy debate got a lot more concrete this week. At its developer conference on June 8, Apple unveiled Siri AI, the rebuilt assistant powered by Google’s Gemini, and confirmed that it will not ship on iPhone or iPad in the EU. The European Commission rejected Apple’s proposed Digital Markets Act compliance mechanisms, including an 18-month phased rollout. Whatever one thinks of the underlying regulatory dispute, the practical result is (maybe not so) unusual: a cutting-edge capability will be reaching users around the world this autumn, before an arrival date in Paris or Madrid is even in sight.
Meanwhile, Brussels’ other hand is writing cheques: the Chips Act 2.0, published June 3, proposes accelerated permitting and a potential €50bn leading-edge semiconductor fab (fabrication plant), with one option under discussion being a one-nanometre facility in Germany by 2035 (not a typo). The pushback came from a Dutch company you may never have heard of, but can no longer afford to ignore. ASML chief executive Christophe Fouquet, whose machines are required for manufacturing every advanced AI chip on earth, warned the Commission against steering strategic projects itself, arguing that such projects must answer to the needs of industry. The warning carries weight: 99 per cent of ASML’s sales are outside Europe.
Between kick-off and lift-off
While Mexico and South Africa kick off the FIFA World Cup tonight, Elon Musk, who is originally South African, has planned his own lift-off: SpaceX sets a price on its shares after the closing bell today and begins trading tomorrow on Nasdaq under the ticker SPCX, targeting $75bn at a $1.75tn valuation, which would make it the largest IPO in market history. It is the first in the trillion-dollar queue: Anthropic and OpenAI both filed confidential prospectuses this month, and analysts estimate the three could seek over $200bn from a market that funded roughly $45bn of US listings in ALL OF LAST YEAR.
A note to our readers
Follow the Token, our 60-minute map of the AI economy, travels with us to Paris next week for a private session alongside VivaTech, where we will also be gathering material for the coming editions. The Barcelona executive briefing follows later this month (seats are limited, register through the link), and the online edition arrives in the second half of July. The longer formats, a half-day executive brief and sector-specific masterclasses, are taking shape; reply or send us a message if you would like an early word.
This week Barcelona hosted a visitor who has already been featured in The Compute and whose institution has outlasted every parliament on the continent. Pope Leo XIV spent the week in Spain and yesterday evening inaugurated the Sagrada Família’s final tower, completing Gaudí’s basilica 144 years after construction began, on the centenary of the architect’s death. He arrived fresh from publishing Magnifica Humanitas, his first encyclical, on safeguarding the human person in the age of artificial intelligence. The Vatican, of all institutions, has published its AI position before most legislatures.
Wirelessly yours,
Ziad Matar
Co-founder, Tokenando
Editor-in-chief, The Compute
After Hours.
This issue of The Compute was written over a mix of 90s music, brought to us this week by The Corrs, which kept us Breathless just like this week did. Office Romance provided some mild comic relief, proving that fifty-something executives can still deliver a happy ending. Meanwhile, we are still watching the Kim K memes from the Monaco Grand Prix while Barcelona gets ready for its own F1 race this weekend.
The essay: “Policy on the AI Exponential” by Dario Amodei, June 2026.
The Encyclical: Magnifica Humanitas, May 2026.
Read Issue VI of The Compute here / https://tokenandoai.substack.com/p/the-essay-or-the-movie?triedRedirect=true