Anthropic’s public calling card is “AI safety first.” Because it makes safety its brand, every controversy it faces gets extra scrutiny.
This piece lays out its risks: the copyright lawsuits, the standoff with the U.S. Department of War, the consistency of its safety pledges, and its financial structure. It marks clearly which issues are settled and which are still open. If you want to get to know the company first, start with What kind of company Anthropic is.
Copyright: One Settlement, Several Cases Pending
Where training data comes from is a legal problem shared by every large AI company.
Books: In 2025, a U.S. federal court ruled that the AI-training use in this case was fair use, but downloading pirated books to build a permanent library was not. Anthropic settled the piracy part for $1.5 billion, covering more than 480,000 books at about $3,000 each. It is the largest copyright settlement in U.S. history, and the court gave final approval on July 20, 2026. “Training on books” won partial support; “where the books came from” came at a steep price.
Music: Publishers including Universal Music and Concord have sued since 2023, alleging that Claude misused copyrighted lyrics. Another case filed in January 2026 expanded to about 20,000 works and seeks about $3 billion. In August 2026, music publishers owned by Sony and Warner also sued, naming not only the company but also CEO Dario Amodei and co-founder Ben Mann. As of October 2026, none of these cases has been decided yet; outcomes and amounts are open.
The Defense Dispute: A Safety Line Taken to Court
In early 2026, the U.S. Department of War (formerly the Department of Defense) wanted to loosen some limits on how Claude could be used. The dispute centered on uses like mass surveillance and fully autonomous weapons. CEO Dario Amodei said publicly that he would not accept removing the relevant safeguards, and the two sides deadlocked.
The Department of War then labeled Anthropic a “supply-chain risk” under two different legal authorities. Anthropic sued and won one case and lost one: on August 27, 2026, the federal district court for the Northern District of California struck down one designation; on September 25, the federal appeals court in Washington, D.C. upheld the other exclusion by a 2–1 vote. Anthropic said it is weighing its options, including further review.
Supporters say this proves its safety pledges can withstand government pressure. But if a settlement or concessions follow, whether that stance holds still needs watching.
Is the Safety Brand Consistent?
When safety is your calling card, the worst charge is “saying one thing and doing another.” TIME reported that Anthropic loosened an early flagship safety pledge, roughly “don’t train a model unless you can guarantee in advance that the safeguards are adequate.” Anthropic responded with a new Responsible Scaling Policy, stressing more transparency and outside review.
Critics say the loosening lined up with commercial expansion, as if safety were making way for growth. Anthropic says the new framework is more practical and easier to enforce. Both accounts exist, and there is no settled answer. The structural roots of this tension are covered in Why Anthropic is a public benefit corporation.
Finances: Big Losses, Bigger Commitments
As of October 2026, Anthropic’s prospectus is not public, but Reuters saw it and reported the figures below (audited financials are not yet public):
| Item | Figure |
|---|---|
| 2025 revenue | Nearly $4.6 billion |
| 2025 operating loss | More than $8 billion |
| 2025 net loss | About $42 billion, of which about $34 billion is non-cash accounting charges |
| Cloud, compute, and infrastructure obligations | $518 billion |
Separately, according to Bloomberg, Q2 2026 revenue topped $11.5 billion and adjusted operating income turned positive (excluding stock-based compensation).
The risk sits in two places. First, the price war: OpenAI, Google, and the open-weight camp are all pushing prices down, which squeezes margins over the long run. Second, the $518 billion in obligations is far larger than current revenue; if growth slows, those fixed costs become a burden (the compute strategy is covered in Anthropic’s compute gamble).
Penchan’s summary
Anthropic’s biggest risk comes, in a way, from its biggest selling point. When “safety” is your core brand, the bar is set especially high: every loosened pledge, every lawsuit, every standoff with the government gets put under a magnifying glass. Hold the line, and the brand becomes more convincing; fail, and the backlash hits harder than it would for an ordinary company.
Huge losses and long-term compute contracts are risks shared by frontier AI companies, and they still need watching.
Further reading: What kind of company Anthropic is, Anthropic’s valuation and IPO, Why it’s a public benefit corporation.