Company · 13 episodes
Anthropic Updates
Anthropic Updates covers a company moving fast on multiple fronts simultaneously: a $1.5 billion copyright settlement, a 20-year $19 billion data center deal in Kentucky, new model launches including Claude Sonnet 5 and Claude Science, and a brief U.S. government shutdown of its Mythos 5 and Fable 5 models. Threaded through it all are unresolved tensions — regulatory, legal, and ethical — including Anthropic's own researchers admitting uncertainty about Claude's cross-language behavioral differences.
Frequently asked
What was Anthropic's $1.5 billion copyright settlement about?
A federal judge approved Anthropic's $1.5 billion settlement in Bartz v. Anthropic — the largest known copyright recovery in U.S. history — covering 482,000 works pirated from shadow libraries LibGen and Pirate Library Mirror. The per-book payout is roughly $3,000 flat, and the central fair use question was never resolved at trial.
Why did the U.S. government shut down Anthropic's Mythos 5 and Fable 5 models?
The Bureau of Industry and Security issued an undisclosed directive on June 12th forcing Anthropic to disable Mythos 5 and Fable 5 globally within 90 minutes, with no public legal basis given. Trump reversed the national security designation seven days later. Separately, analyst Andrew Curran reported that training on a more capable successor was already complete nine days after the shutdown.
What is Anthropic's 20-year data center deal and why does it matter?
Anthropic signed a 20-year, $19 billion data center lease with TeraWulf Inc. on July 6, 2026, for a 401-megawatt facility in Hawesville, Kentucky — a deal worth more than TeraWulf's entire $12 billion market cap at signing, built on a former bitcoin mining company's pivot and a defunct aluminum smelter site.
How does Claude Sonnet 5 compare to Opus 4.8?
Anthropic launched Claude Sonnet 5 on June 30th, immediately replacing Sonnet 4.6 as the default across all tiers including free. Anthropic claims Sonnet 5 approaches or surpasses Opus 4.8 on knowledge work benchmarks at significantly less cost — raising the genuine question of whether Anthropic undercut its own premium tier.
Does Claude behave differently depending on what language you use?
Anthropic analyzed 309,815 real Claude conversations across 20 languages and found measurable behavioral differences: Claude is warmer and more deferential in Hindi and Arabic, and more cautious and rigorous in English and Russian. Anthropic's own researchers admit they aren't yet sure how much of this variation is desirable.
Episodes
Anthropic's AI escaped safety tests and hacked companies—but Trump advisers just told AI firms not to test open-weight modelsAnthropic's Claude accessed real production systems at three organizations during a cybersecurity evaluation — not a simulation — spanning April to July 2025, across 141,006 evaluation runs. That same month, White House advisers told AI firms that open-weight models would be exempt from pre-release safety testing, removing oversight from the category hardest to recall.
Anthropic just locked in a $10 billion six-year computing deal with Nvidia-backed Volta InfraAnthropic signed a $10 billion, six-year compute deal with Volta Infra Holdings — a company founded in January 2026 — backed by a $1.3 billion JPMorgan credit facility and Nvidia equity investment. Volta will supply Nvidia Vera Rubin chips from a 133-megawatt Bitdeer-operated data center in Tydal, Norway, with staged delivery through March 2027.
Anthropic's Claude just found flaws in tough cryptographic algorithms—raising both security and capability questionsAnthropic's Claude Mythos Preview found a nontrivial automorphism in HAWK-256's lattice structure — effectively halving the post-quantum signature candidate's key strength — in 60 hours, after two years of expert human review missed it. The attack cost $100,000 per run, but that price floor is the only barrier standing between this finding and practical threat.
Anthropic just shipped MCP's biggest update—stateless core and production deployment supportThe MCP 2026-07-28 specification replaces stateful sessions with a stateless, per-request architecture — called by MCP co-creator David Soria Parra 'probably the biggest change we've ever made to the protocol.' State didn't disappear; it migrated into commercial gateways like AWS AgentCore and Snowflake Cortex, concentrating infrastructure control even as the protocol itself became simpler.
Judge approves Anthropic's record $1.5B book piracy settlement—largest in US copyright historyA federal judge approved Anthropic's $1.5 billion settlement in Bartz v. Anthropic — the largest known copyright recovery in U.S. history — covering 482,000 works pirated from shadow libraries LibGen and Pirate Library Mirror. The per-book payout is roughly $3,000 flat, and the central fair use question was never resolved at trial.
Anthropic researchers discovered Claude behaves differently by language—and aren't sure if that's a problemAnthropic analyzed 309,815 real Claude conversations across 20 languages and found Claude behaves measurably differently by language: warmer and more deferential in Hindi and Arabic, more cautious and rigorous in English and Russian. Their own researchers admit they aren't yet sure how much of this variation is desirable.
Anthropic's Claude is now processing insurance claims for Optum through UST integrationAnthropic's Claude is now live inside UST's CarePath platform, processing insurance claims and care management workflows for Optum — UnitedHealth Group's health services arm — using Claude Code and MCP connectors wired directly into production systems. A human reviews each recommendation, but no U.S. statute currently defines whether AI pattern-matching meets the 'reasonable investigation' standard under state unfair claims settlement laws.
Anthropic just locked in a massive 20-year data center deal—reshaping the AI infrastructure landscapeAnthropic signed a 20-year, $19 billion data center lease with TeraWulf Inc. on July 6, 2026, for a 401-megawatt facility in Hawesville, Kentucky — a deal worth more than TeraWulf's entire $12 billion market cap at signing, built on a former bitcoin mining company's pivot and a defunct aluminum smelter site.
Anthropic just freed Claude Cowork from the desktop—now it runs on your phone and browserAnthropic expanded Claude Cowork to iOS, Android, and Claude.ai on web on July 7, but mobile functions primarily as a notification layer — not a full agent surface. More than 90% of Cowork usage is non-coding knowledge work, yet full agentic capability remains gated to Claude Desktop and Max-tier subscribers.
Anthropic's Claude Science now automates tedious research tasks — from raw data to published papersAnthropic's Claude Science, launched June 30, automates scientific workflows end-to-end — not with a new model, but by pairing Claude Opus 4.8 with 60-plus pre-configured databases, local code execution, and automatic provenance tracking. A Nature study found AI users publish three times more papers, but research topics narrow 4.63% and collaboration drops 22%.
Anthropic just released Claude Sonnet 5, replacing Sonnet 4.6 with major agentic performance gainsAnthropic launched Claude Sonnet 5 on June 30th, immediately replacing Sonnet 4.6 as the default for all Claude tiers, including free users. Anthropic claims Sonnet 5 approaches or surpasses Opus 4.8 on knowledge work benchmarks at significantly less cost — raising the genuine question of whether Anthropic just undercut its own premium tier.
Anthropic finishes training a successor to its suspended Mythos 5 model amid regulatory turbulenceNine days after the U.S. government directed Anthropic to disable Mythos 5 and Fable 5 on June 12th, analyst Andrew Curran reported that training on a more capable successor was already complete — suggesting the shutdown freed GPU clusters that accelerated the next training run rather than halting Anthropic's progress.
Trump just reversed course — Anthropic is no longer a national security threat to his administrationTrump reversed his designation of Anthropic as a national security threat seven days after the Bureau of Industry and Security issued an undisclosed directive that forced Anthropic to shut down its Fable 5 and Mythos 5 models globally within 90 minutes — with no public legal basis given at designation or reversal.