It is Thursday, June 4, 2026. Four stories broke or developed significantly today and each one of them is worth sitting with. Here they are. Claude is now writing the code that trains Claude. Anthropic said so publicly today and paired it with numbers that are hard to sit with. Meanwhile SpaceX opened its roadshow, OpenAI pushed deeper into drug discovery, and Florida filed the first state lawsuit to name Sam Altman personally. Here is what happened on June 4, 2026.
Anthropic says Claude is now writing its own future — and the numbers are unsettling
The biggest story of the day did not come from a product launch. It came from Anthropic’s own research institute, which published a report this morning disclosing just how much of Anthropic’s production code is now authored by Claude.
The answer, as of May 2026: more than 80 percent. Before Claude Code launched in research preview in February 2025, that number was in the low single digits. The typical Anthropic engineer now ships eight times as much code per day as they did in 2024. On a specific code-speedup benchmark where Claude is handed training code and asked to make it run faster, the Mythos Preview model hit a 52x speedup. Skilled human engineers hit 4x on the same test.
The report also disclosed that when research sessions went off track, Claude proposed a better next step than the human took 64 percent of the time. On the most open-ended engineering tasks, Claude’s success rate jumped from 26 percent to 76 percent in six months.
Anthropic is explicit that this is not yet recursive self-improvement — the threshold where AI autonomously designs and builds its own successors without meaningful human direction. But the company says it could arrive sooner than most people expect, and the framing of the report makes clear this is a deliberate disclosure, not a casual aside.
Read the original report: Anthropic Institute: When AI builds itself
Read the coverage: BeInCrypto breakdown | Investing.com analysis
SpaceX kicked off the roadshow for what could be the largest IPO in history — today
The SpaceX investor roadshow officially launched this morning, June 4, with the company releasing its full roadshow presentation and opening retail share access through Fidelity. Pricing is set for June 11. Trading begins June 12 on Nasdaq under the ticker SPCX.
The target: raise $75 billion at a valuation of $1.75 trillion. For context, Saudi Aramco’s 2019 IPO — previously the largest in history — raised $29 billion. SpaceX has reserved up to 30 percent of the offering for retail investors, an unusually high portion. Fidelity reduced its access threshold to any brokerage account holding $2,000 or more.
The company disclosed 2025 revenue of $18.7 billion, driven primarily by Starlink’s satellite broadband business, which now serves more than nine million subscribers. Elon Musk will not sell a single share. The dual-class structure gives him full voting control after listing.
Polymarket currently gives a 94 percent probability to the IPO completing in the June window, with 72 percent odds it clears a $2 trillion market cap on debut. This is the most consequential public offering in AI-adjacent infrastructure since the sector emerged.
Read the coverage: Full roadshow timeline and retail access details | Reuters on the retail investor strategy
OpenAI upgraded GPT-Rosalind and opened it to drug researchers worldwide
OpenAI published a substantial update to GPT-Rosalind yesterday, its domain-specific model purpose-built for life sciences research. The update combines GPT-5.5’s agentic coding capabilities with stronger performance in medicinal chemistry, genomics, and wet-lab troubleshooting — the three areas where pharmaceutical research generates the most compute-intensive workloads.
Three new benchmarks show the updated model outperforms GPT-5.5 across all tested domains while using fewer tokens. In genomics specifically, GPT-Rosalind completes long-horizon analyses using 31 percent fewer tokens than GPT-5.5. New Codex plugins connect researchers to more than 50 scientific databases directly within the model interface.
The research preview is now open to eligible organizations worldwide for the first time. Novo Nordisk joined as a named partner, with its researchers using the model to analyze complex datasets and connect evidence across genomics, transcriptomics, and structural biology. The broader Rosalind Biodefense program, which provides government and public-health access for epidemiological modeling and medical countermeasure development, was also expanded.
Read the original announcement: OpenAI: Introducing new capabilities to GPT-Rosalind
Read the coverage: TechTimes on the genomics efficiency gains | Digit on the LifeSciBench benchmark results
Florida became the first state to sue OpenAI — and named Sam Altman personally
Florida Attorney General James Uthmeier filed a lawsuit against OpenAI and CEO Sam Altman on Monday, making Florida the first state to take legal action against the company. The suit names Altman personally, accusing him of being “very central” to pushing product features that the state says caused measurable harm.
The complaint covers a wide range of allegations: that OpenAI suppressed internal safety warnings, marketed ChatGPT to minors without adequate parental controls, collected children’s data without meaningful consent, caused behavioral addiction and cognitive harm, and provided information that was cited in connection with a shooting at a Tallahassee university. The suit invokes Florida’s product liability and unfair trade practices laws and seeks civil penalties plus a court order blocking certain data collection from users under 13.
OpenAI responded that it has “put in place industry leading protections and policies” for minors and that it is “committed to getting this right.” The lawsuit follows Florida’s investigation into OpenAI that was announced in April, which itself came after the Tallahassee shooting.
Other states have followed a similar playbook against AI chatbots. Kentucky sued Character.AI in January. Utah sued Snap over its AI chatbot in June. Florida is the first to go after OpenAI specifically, and the first to name a CEO as a personal defendant in an AI safety case.
Read the coverage: CBS News | CNN on the allegations | Fortune on the complaint details
That is June 4, 2026. Claude is writing its own code, SpaceX opened its books to the public, OpenAI went deeper into drug discovery, and the first state lawsuit naming an AI CEO by name is now in the courts. See you tomorrow.