AI This Week
SpaceX (SPCX) shares jumped 17% after the newly public aerospace giant announced a $60 billion acquisition of Anysphere, the startup behind AI coding assistant Cursor. The move is seismic. The announcement came just days after SpaceX completed its IPO and began trading on Nasdaq under the ticker "SPCX." SpaceX topped Amazon and Microsoft by market cap, making it the fourth most valuable company in the U.S. Under the deal terms, SpaceX entered into a merger agreement with Anysphere and subsidiary X67, which will merge into Cursor — surviving as a wholly owned SpaceX subsidiary. Founded in 2022, Cursor quickly emerged as a major player in AI-powered software development, helping developers generate, edit, and review code. In November, Cursor crossed $1 billion in annualized revenue. However, Cursor's market share had declined from 41% in June 2025 to about 26% in May. If the deal is not completed, SpaceX agreed to pay a termination fee of $1.5 billion plus $8.5 billion in computing resources. The transaction is expected to close in Q3 2026.
Meta launched Facebook AI Mode, a brand-new search experience that uses Meta AI to synthesize conversational answers directly from public posts. The feature transforms the standard search bar into a conversational tool that answers questions by mining public Group discussions, Reels, and Marketplace data. AI Mode runs on Meta's Muse Spark model — ask it a question and it synthesizes a response from what real Facebook users have written, surfacing forum-style advice, event recommendations, or product tips. Users can also ask follow-up questions, making the search experience feel more like interacting with a chatbot than using a conventional search engine. But the launch isn't without controversy. AI Mode draws from whatever users have publicly written — including misinformation — putting it at a structural disadvantage versus competitors like Google's Search Generative Experience or Perplexity. Meta's approach hinges on a simple trade: if you posted it publicly, it's now training data — with no opt-out mentioned in the announcement.
Anthropic has pulled the plug on two of its most powerful AI models. The company disabled access to Fable 5 and Mythos 5 to comply with an export control directive from the U.S. government that cited "national security authorities." The directive arrived at 5:21 p.m. ET on Friday. Both models had launched just three days earlier, on June 9. The directive covers not just people located outside the U.S., but also any foreign national inside the U.S., including Anthropic's own non-citizen employees — leaving the company no choice but to disable the models for all users. Officials told Anthropic the decision came after the government learned of a technique to bypass Fable 5's safeguards, which were designed to prevent access to the powerful cybersecurity abilities of Mythos, the underlying model on which Fable 5 is built. The letter did not provide specific details of the national security concern. Access to less powerful Claude models, including Claude Opus 4.8, was not affected. This appears to be the first government-forced takedown of a publicly deployed frontier AI model.
SpaceX priced its IPO Thursday night at $135 a share, debuting on the Nasdaq under the ticker SPCX. The offering raised more than $75 billion, easily outranking previous record-holder Saudi Aramco's $29.4 billion debut in 2019. The company placed in the top 10 of Wall Street's biggest companies with a valuation of just under $1.8 trillion — ahead of Tesla, Meta, and Walmart. The record IPO is a testament to Musk's continued investor support, with the offering more than four times oversubscribed. SpaceX planned to offer 30% of the IPO to everyday investors through retail brokerages such as Fidelity, E-Trade, Charles Schwab, and SoFi. A successful debut could make Musk history's first trillionaire. While SpaceX is growing fast — revenue hit $18.7 billion in 2025 — it is also losing money, posting a net loss of $4.9 billion, mainly on AI spending. SpaceX is also the first among leading AI giants to go public, with OpenAI and Anthropic both recently filing initial documents with regulators.
Microsoft, despite being one of Anthropic's biggest backers, is quietly restricting internal use of Claude Fable 5. The decision stems from Anthropic's updated data retention requirements, which have raised red flags inside Microsoft over how employee and customer information might be handled. Claude Fable 5 is not available in the model picker used by Microsoft employees for internal GitHub Copilot, while other Claude models remain available internally under Zero Data Retention (ZDR) rules. Claude Fable 5 was unveiled as Anthropic's first "Mythos-class" model and rolled out to GitHub Copilot and Foundry customers almost immediately. Anthropic ties Claude Fable 5 to its safety classifiers so that prompts and outputs are retained for 30 days, and content flagged as violating Anthropic's usage policy can be kept for up to two years. Microsoft has told employees that its legal teams are evaluating Anthropic's retention changes because of concerns about customer data and confidential information.
Anthropic is rolling out a public version of its Mythos AI model, but with guardrails barring its use in risky areas such as cybersecurity, after a preview earlier this year sent shockwaves globally with its ability to find software flaws. The new Claude Fable 5 is the most powerful model Anthropic has ever made for wider use, touting its performance in software engineering and analytics. In high-risk areas like cybersecurity, biology, chemistry, and distillation, the model blocks responses and falls back to Claude Opus 4.8. The big question is whether the guardrails are strong enough to withstand jailbreaking attempts — Anthropic says it "extensively" tested the model with hackers who tried to bypass its safeguards, and none were successful. Anthropic has so far limited Mythos access to about 200 organizations, including the U.S. government, under the Glasswing program, after announcing in April that Mythos had uncovered thousands of software vulnerabilities. Anthropic also rolled out an upgraded Mythos 5 to select customers, which "has the strongest cybersecurity capabilities of any model in the world." Offering its capabilities more widely may allow the $965 billion company to extend the momentum that has powered its valuation above rival OpenAI. Pricing on both models is $10 per million input tokens and $50 per million output tokens.
Argentina is making a stunning legal bet on artificial intelligence. President Javier Milei published an op-ed in the Financial Times pitching Argentina as a haven for tech giants, vowing to let AI develop freely "without the deadly hand of premature and poorly understood regulation." The three pillars of his framework are a commitment to keep AI unregulated, the creation of "a new corporate category in Argentine law: the non-human corporation," and a low corporate tax rate. If approved, the initiative would grant legal status to "non-human companies" managed exclusively by AI, and would also create the legal entity of decentralized organizations. Legal experts say this would make Argentina the first in the world with such legislation, enabling AI agents to execute operations, enter into contracts, and move funds independently — with their own legal entity and limited liability, without requiring a human behind them. Argentina's Congress is weighing the world's first law for non-human corporations, as Peter Thiel relocates to Buenos Aires while Milei courts AI firms with radical deregulation. Politicians and tech experts have expressed concern over the implications, warning of AI entities operating with no regulations and low taxes. Critics argue it would create "programmed impunity: human gains, social harm and responsibility shifted onto machines."
Elon Musk's SpaceX has entered into a multi-year cloud services agreement with Google, valued at $920 million per month starting from October 2026. SpaceX outlined the terms in a regulatory filing, under which it will provide Google with approximately 110,000 Nvidia GPUs, CPUs, memory, and associated components. Payments will begin at a reduced rate during a ramp-up period through September before reaching the full monthly fee from October until June 2029. If the contract runs its full course, SpaceX stands to collect roughly $30 billion. A Google representative described the deal as a result of unexpected demand for its recently launched AI products. It follows a similar deal with Anthropic, which agreed to pay SpaceX $1.25 billion per month through 2029 for compute from its Colossus 1 data center near Memphis, Tennessee. If SpaceX fails to deliver the committed GPUs by September 30, Google may terminate the contract or accept fewer GPUs at a reduced fee.
OpenAI has overhauled ChatGPT's memory with a breakthrough system it calls "dreaming." The AI enters a state where it puts down its work and reviews all past chats — where it went wrong and where it got things right. The results are dramatic. OpenAI claims the new feature bumps ChatGPT's factual recall successes to 82.8%, up from 41.5% in 2024. ChatGPT could dream since 2025, but with this latest update, OpenAI is making the AI remember more information about you and your lifestyle. Users can pop into ChatGPT's summary after it finishes dreaming and tweak things that may have been misremembered. The feature is rolling out to all users, including the Free tier, right now.
Meta has repeatedly pushed back plans to release its Muse Spark AI model API to developers, and as of Tuesday had no scheduled launch date, the Wall Street Journal reported. The company postponed the API release from the original target of April to May, citing bugs discovered during testing and the need to build additional infrastructure. Meta first unveiled Muse Spark in April, describing it as the first model designed to narrow the gap with competitors — the initial release from its new Superintelligence Labs. That's a glaring bottleneck for a company spending up to $145 billion on AI infrastructure this year: without an API, Meta can't easily sell access to the model, ceding a lucrative monetization engine to rivals like OpenAI and Anthropic. A Meta spokesperson said the company is already testing the API with some early partners and is looking forward to releasing it this month.