Welcome to AI Decoded, Quick Firm’s weekly publication that breaks down an important information on this planet of AI. You possibly can signal as much as obtain this text each week by way of e-mail here.
Bullies vs. Brains: Anthropic’s new scrape with the Trump administration
Anthropic is in one other battle with the Trump administration, this time over its new Mythos-class fashions, launched final week. On Friday, administration officers panicked after stories that Amazon researchers had tricked Claude Fable 5 into offering cybersecurity data Anthropic had tried to dam. Officers gave the corporate 90 minutes to voluntarily take Claude Fable 5 and Claude Mythos 5 offline.
Anthropic, ready for proof that the fashions had truly been compromised, didn’t instantly comply. The administration then declared the fashions a cybersecurity danger and barred overseas nationals from utilizing them. As a result of Anthropic had no sensible option to restrict entry to U.S. residents solely, it shut the fashions down for everybody.
Because of this, no person can use Claude Fable 5 or Claude Mythos 5, doubtless essentially the most highly effective publicly obtainable AI fashions on this planet. Which will maintain unhealthy actors from exploiting them, however it additionally prevents cyberdefense researchers and software program firms from utilizing them to cease cyberattacks. Even overseas nationals who work at Anthropic are actually barred from utilizing the fashions.
The proof for the panic seems skinny. The cybersecurity knowledgeable Katie Moussouris reviewed the Amazon researchers’ report and told The Atlantic that Fable had refused a direct request to assessment insecure code for safety flaws, however complied when requested to “repair this code,” adopted by extra handbook steps. The White Home had additionally reportedly heard {that a} China-linked group had gained entry to Mythos, however the authorities introduced no proof, and Anthropic disputed the declare.
This marks the second main dustup between the Trump administration and Anthropic, and the primary was simply as foolish. The Division of Protection banned using Anthropic fashions by protection contractors large and small after Anthropic caught agency to its guidelines barring use of the expertise for concentrating on autonomous weapons and surveilling U.S. residents. Some feared that after being branded a “provide chain danger,” Anthropic would see its enterprise nosedive. As an alternative, the nice vibes round Anthropic elevated and its success with enterprises accelerated. Now an IPO is on the horizon.
That the administration would single out an AI firm it branded “woke” for punishment is hardly shocking. However the Mythos ban additionally exposes a deeper drawback: The administration is making use of a particular, advert hoc customary to Anthropic’s fashions that no different U.S. mannequin faces. That’s partly as a result of, beneath “AI czar” David Sacks, the administration has never established a transparent danger framework for evaluating frontier fashions like Mythos. The result’s perverse: A few of America’s strongest AI fashions have been taken out of motion, whereas competing Chinese language fashions stay unencumbered by U.S. regulators and proceed to enhance.
To make certain, Anthropic deserves critical scrutiny. However the White Home battle over Fable and Mythos has blurred extra vital questions on how the corporate truly governs its fashions.
Within the Fable 5 launch notes, Anthropic stated it will start retaining consumer prompts for 30 days to assist detect misuse, and the info wouldn’t be used for coaching. However the coverage nonetheless raises apparent privateness issues. The corporate additionally stated it will silently degrade Fable 5’s output if it detected that the mannequin was getting used to coach rivals. That lack of transparency was troubling, and Anthropic has since halted the apply.
To skeptics, this was all proof that Anthropic will not be actually so completely different from its rival, OpenAI. Ben Thompson, the tech analyst who runs the Stratechery publication and podcast community, argued that Anthropic’s willingness to degrade the output of its fashions to thwart rival mannequin makers factors to one thing darker: that Anthropic “doesn’t suppose that anybody else apart from them ought to even be making frontier LLMs.” However isn’t it doable that Anthropic actually was simply attempting to forestall different mannequin makers from ripping off its personal mannequin’s intelligence, even when it selected a sneaky and ill-advised means of doing it? It’s fairly a leap to collect from that that Anthropic believes it’s the one firm able to constructing frontier giant language fashions.
Thompson means that silently dumbing down mannequin output is akin to coaching its fashions to not help customers, even protection customers, in concentrating on autonomous weapons or surveilling U.S. residents. “What this degradation represented was each the aptitude and willingness of Anthropic to silently alter its fashions to attain its coverage preferences,” he writes. “In different phrases, Anthropic willfully validated a few of its critics’ worst fears when it comes to being a provide chain danger.”
However doesn’t Anthropic have a respectable curiosity in setting limits on how its personal fashions are used, even in protection? For instance, the corporate defined that it prohibited using its fashions for home mass surveillance partially as a result of present U.S. regulation “doesn’t but account for the breadth and pace of mass surveillance that AI might allow.”
Why SpaceX is shopping for Cursor simply days after its IPO
Like Mark Zuckerberg, Elon Musk badly desires to have a horse within the race for the largest and finest general-purpose AI mannequin. Proper now Meta and xAI (now a part of SpaceX) aren’t aggressive with Anthropic, OpenAI, and Google Deepmind, however they need to construct or purchase the items they should compete. An AI-assisted coding software was an enormous lacking piece for xAI, therefore SpaceX’s acquisition of Cursor and its individuals. And laptop code is a key ingredient for coaching AI fashions, so Cursor’s coding information might find yourself growing the standard of xAI’s Grok fashions and chatbot.
xAI has been codeveloping fashions with Cursor since April, a partnership that gave SpaceX the choice to purchase the startup for $60 billion. SpaceX will reportedly pay Cursor in post-IPO inventory, not money. Cursor has been rising quickly since 2022, reaching greater than one million paying customers and $4 billion in yearly recurring income in 2026.
Will the deal make xAI a top-tier AI lab? “I’m a bit skeptical on that, as xAI’s fashions have lagged considerably, and one robust coding mannequin doesn’t shut that hole,” writes the Pitchbook analyst Franco Granda in a brand new report.
The deal is sensible in some methods. Cursor will get locked-in entry to SpaceX’s Colossus supercomputer to coach new fashions, which might speed up its competitiveness relative to Anthropic’s Claude Code and OpenAI’s Codex coding instruments. Cursor is an AI-native software program software the place programmers write, edit, run, check, and debug code. It’s attempting to change into a spot the place builders supervise fleets of coding brokers reasonably than simply work with a single agent to put in writing code themselves.
Cursor might present SpaceX with a pathway to promote entry to its fashions and instruments to enterprises. “I consider what issues extra to SpaceX is [Cursor’s] distribution of over one million day by day energetic customers . . . a distribution footprint xAI has not been in a position to construct by itself,” Granda writes. That’s vital as a result of a lot of SpaceX’s pitch to buyers was its plan to change into primarily an enterprise AI firm.
ChatGPT remains to be the chatbot king, however rivals are gaining floor
OpenAI’s ChatGPT stays the most-used AI chatbot, however Anthropic’s Claude and Google’s Gemini are catching up, in accordance with a new report from Sensor Tower, a digital market intelligence agency.
ChatGPT grew 4.8% from 1.05 billion distinctive month-to-month customers in December 2025 to 1.1 billion customers in Might, the report says. However Gemini is gaining floor. Thanks partially to a big Android consumer base and deep integrations inside Google’s ecosystem of merchandise, Gemini has grown from 533 million customers in December 2025 to 662 million in Might 2026. (That’s a 24% uptick, for these conserving rating.)
Claude, in the meantime, has skilled fast development for the reason that begin of 2026. In December 2025 the chatbot had simply 60 million month-to-month distinctive customers throughout cellular and desktop, however that quantity elevated fourfold to 245 million by Might 2026.
Sensor Tower measures distinctive chatbot customers in 25 world markets throughout cellular apps, cellular internet, and desktop internet, so it could have essentially the most full image obtainable of chatbot market share.
Extra AI protection from Quick Firm:
- Inside Fei-Fei Li’s $1 billion new AI company, World Labs
- The World Cup security buildout won’t end when the games do
- This hidden Gemini feature uses AI to teach you to be a tech savant
- AWS says AI agents can work on their own. It’s also building tools to keep them in line
Need unique reporting and development evaluation on expertise, enterprise innovation, future of labor, and design? Sign up for Quick Firm Premium.
