Anthropic’s public calling card is “AI safety first,” but no company can have only a polished side. Making safety its brand, if anything, puts every one of its controversies under especially magnified scrutiny.
This piece lays out Anthropic’s risk list honestly: the copyright lawsuits, the tug-of-war with the Department of War, the doubts over the consistency of its safety pledges, and a financial structure with no visible floor. It tries to present both sides side by side and to mark clearly which questions are not yet settled. If you want to get to know the company first, you can start with What kind of company Anthropic is.
To set the tone in one sentence: for a company whose calling card is safety, the real risk hides in whether a gap opens up between its conduct and that calling card.
Copyright: One Settlement, One Still Unresolved
Where training data comes from is a legal minefield shared by all large AI companies, and Anthropic is standing on it too, mainly along two fronts.
Book copyright: In 2025, a U.S. federal court issued a partial ruling holding that training AI on legally obtained books can count as fair use, but the portion obtained from “pirated libraries” does not. For the piracy portion, Anthropic later agreed to pay about $1.5 billion to settle — an amount equivalent to roughly 500,000 works at about $3,000 each — setting the record for the largest copyright settlement in U.S. history; the settlement received preliminary court approval in September 2025, with the final-approval hearing scheduled for mid-2026. “Training on books” won partial support, while “where the books came from” came at a steep price.
Music copyright: Music publishers including Universal Music Group (UMG) and Concord have sued Anthropic since 2023, alleging that its models improperly used copyrighted lyrics. In early 2026 the litigation expanded, with claims covering roughly 20,000 works and seeking about $3 billion in damages. This is a lawsuit that is still undecided — the final outcome and amount remain unset, and it cannot be treated as established fact.
The Defense Dispute: A Tug-of-War Over the Safety Line
In early 2026, a public disagreement between Anthropic and the U.S. Department of War pushed its safety stance onto a real-world exam.
At the heart of the dispute is whether Claude can be used for certain highly sensitive purposes. The Department of War’s starting point is national security and operational need, and it wanted to loosen some usage restrictions; CEO Dario Amodei publicly stated that he could not accept demands to remove the relevant safety safeguards, and the two sides deadlocked for a time. Each side has its own position, presented side by side here without taking a side.
How to read this depends on your vantage point. Supporters see it as proof that Anthropic’s safety pledges are “for real” — able to withstand even government pressure; others caution that if a contract is signed or conditions are conceded in the future, whether this stance holds still needs to be tracked. This piece lays both sides’ accounts side by side, reaches no conclusion for either side, and treats subsequent developments as worth watching closely.
The Consistency Tension of a Safety Brand
When you make safety your calling card, the thing you fear most is being caught “saying one thing and doing another.” Anthropic is facing this kind of question.
Some media (such as TIME) have reported that Anthropic loosened a tougher flagship safety pledge from its early days — for example, the high bar of “not training before it can guarantee in advance that the safety safeguards are sufficient.” In response, Anthropic pointed to a new version of its Responsible Scaling Policy, emphasizing that the new version adds more transparency and external-review design.
Critics argue that the timing of loosening the pledge overlaps with commercial expansion, which inevitably raises the suspicion that safety is making way for growth; Anthropic’s side maintains that the new framework is a more pragmatic, more executable version. Both accounts exist, and which one has the better case has no settled answer yet — it is a key observation point for testing the company’s brand consistency. The structural roots of this tension are laid out in Why Anthropic is a public benefit corporation.
Commercial Pressure: Pricing, Gross Margin, and Cash Burn
Setting the legal and safety controversies aside and looking purely at the business, Anthropic has its hidden worries too.
The inference cost of AI models is not low, and the pricing competition among rivals is fierce — OpenAI, Google, and the open-weight camp are all driving prices down. This means Anthropic’s gross margin will face downward pressure over the long run. The problem is that, as a non-public company with no audited financials, its gross margin, the scale of its losses, and how fast it burns cash are all invisible to outsiders.
Layer on top of that the sky-high long-term compute contracts it has taken on — fixed-cost obligations that far exceed current revenue (that thread is laid out in Anthropic’s compute gamble) — and whether it can turn a steady profit over the long term is a question that honestly has to be answered “not yet confirmable.”
The Parts Not Yet Laid Open
When discussing risk, the worst mistake is to dress up an “unknown” as “known bad news.” Here are several parts that currently can’t be seen, marked clearly:
- Gross margin and cash burn: Gross margin, losses, and the rate of cash consumption are all undisclosed, so outsiders can’t assess long-term sustainability. There are scattered reports of nearing breakeven or even turning a profit, but they are single-source and unverified, and this article does not treat them as settled.
- Litigation outcomes: The music-publisher suit is still undecided, with the amount unset; the book-copyright case has been settled for about $1.5 billion, but final approval still awaits confirmation at a court hearing.
- The gray market: A gray market has emerged of agents reselling Claude API access at low prices, accompanied by data-security concerns; this is a reported phenomenon, but its details and scale are hard to verify.
These gaps remind us that, among Anthropic’s risks, some are controversies that have already happened, while others are merely unknowns that are “not yet seen clearly” — and the two should not be conflated.
Penchan’s Take
After going through this risk list, you’ll find Anthropic’s situation a bit peculiar: its biggest risk, to some degree, comes precisely from its biggest selling point.
When a company makes “safety” and “responsibility” its core brand, the outside world holds it to an especially high moral standard. Any loosening of a pledge, any lawsuit, any dispute with the government gets examined under a magnifying glass. For Anthropic this is a double-edged sword: hold the line, and the safety calling card becomes more persuasive; fail to hold it, and the backlash hits harder than it would for an ordinary company.
As for the financial opacity and the cash burn, those are a question mark shared across the entire frontier AI industry — not unique to Anthropic. For anyone trying to understand this company, watching how these risks “evolve” gets closer to the truth than slapping a one-off good-or-bad label on it.
Further reading: What kind of company Anthropic is, Anthropic’s valuation and IPO, Why it’s a public benefit corporation.