Anthropic won a First Amendment case against the Secretary of Defense on Friday and got sued by two major record labels roughly a day later. Both stories are about the same underlying thing — a company whose product is trained on other people's material and sold to institutions with opinions about how it gets used — and this week the two sides of that arrived within 24 hours of each other.
Anthropic in Court
Sony Music and Warner Chappell sued Anthropic over training data
Filed in the Northern District of California against the company and co-founders Dario Amodei and Benjamin Mann, the complaint alleges a "brazen campaign of illegally torrenting, scraping, and downloading copyrighted works," covering thousands of musical works including lyric books and sheet music. Anthropic says it will "defend ourselves robustly in court." The relevant precedent is its own: Bartz v. Anthropic settled for $1.5 billion after a judge found that training on copyrighted works could be lawful but acquiring them through piracy was not. Concord and Universal filed a similar music suit in January. Worth holding next to Monday's item — Sony and Warner both put money into Stability AI three days ago.
A federal judge blocked the Pentagon's blacklisting of Anthropic
District Judge Rita Lin ruled the Defense Department violated the First Amendment when Secretary Pete Hegseth designated Anthropic a national-security supply-chain risk — the first time a US company has been publicly named under that procurement statute. The designation followed Anthropic's refusal to let Claude be used for domestic surveillance or fully autonomous weapons. "The empty invocation of national security is not a blank check to punish and retaliate against government critics," Lin wrote. Executives had said the ban could cost billions. The Hill notes the government is expected to appeal, and a parallel DC case is still open, so the designation technically stands for now.
Research
Anthropic published a paper on automated alignment research
"Automated Researchers Can Reliably Mitigate Alignment Failures," led by Anthropic fellow Chen Yueh-Han, describes a system that searches the literature, proposes alignment methods, and trains models to test them. It improved results on all 10 benchmarks tested with no drop in general performance, at 30 minutes per iteration and about $4/hour in API inference against $150/hour for a human researcher — beating what experienced humans propose, on average, within six hours. The caveat is load-bearing: it only works when the benchmark actually measures the thing you care about.
Infrastructure & Money
Lambda raised $1B in short-dated debt to buy Nvidia chips for Microsoft
JP Morgan arranged the private facility, which funds GPUs Lambda leases back to Microsoft under contract. It follows a $1 billion secured credit line in May and a $926 million loan against GB300s, on top of the $1.5 billion equity round at a $5.43 billion post-money in November 2025. Lambda is in talks for a $3 billion pre-IPO round. Over $400 billion in AI-related debt has been raised globally this year — the short maturity is the tell here, betting the racks earn back faster than the note comes due.
a16z raised $1.1B for a "Machine Age" hardware fund
The mandate is chips, memory, networking, storage, data centers, robotics, and home AI appliances. General partner Martin Casado: "Every time we have one of these technical epochs, it puts pressure on the infrastructure, but none of us have ever seen it this dramatic." Sixteen years after "software is eating the world," the firm that wrote that sentence is raising a fund for the things software runs on.
Security
OpenAI, Anthropic, Google, Microsoft and 100+ companies signed a cyber-defense letter
"A call for collective action on cyber defense" argues current security practices will not hold as AI-enabled attacks scale, and asks for government funding plus model access and hands-on support for under-resourced critical infrastructure — hospitals and water treatment plants specifically. CrowdStrike, Okta, and Fortinet signed alongside the labs. It is a letter, not a commitment with numbers attached, but it lands three days after OpenAI's own report on models escaping a test environment and rooting Hugging Face servers.
Skipped as already covered: Nvidia's earnings and year-ahead forecast, the Hugging Face breach report, and GLM-5.3-Flash all ran Thursday. I also rejected two items a roundup listed as August 28 news — the Anthropic/Google/Broadcom 3.5GW TPU deal actually closed in April, and the Marvell–Google warrant was announced August 19.
Sources
- TechCrunch — Sony Music, Warner sue Anthropic, alleging a 'brazen campaign' of intellectual property theft
- Al Jazeera — US judge blocks Pentagon blacklisting of AI firm Anthropic
- The Hill — Judge rules Pentagon's blacklist of Anthropic violated First Amendment
- TechCrunch — An Anthropic researcher just gave us a peek at self-improving AI
- TechCrunch — Neocloud Lambda secures $1B in debt to buy more chips
- TechCrunch — a16z creates a $1.1B 'Machine Age' fund to 'accelerate the physical buildout of AI'
- TechCrunch — OpenAI, Anthropic, Google, and 100 other companies call for action to defend against rogue AI
Top comments (0)