20 JAN 2026policy
Amodei compares chip exports to selling nuclear weapons.
At Davos, Amodei likened selling advanced chips to China to selling nuclear weapons to North Korea, according to Reuters reporting.
Anthropic’s response / context
The analogy accompanied his opposition to permitting those advanced chip sales.
The important qualification
This is an attributable statement about export policy. It should not be transformed into a claim about the character of Chinese people.
27 JAN 2026copyright
Project Panama cut purchased books apart to digitize them.
Later reporting; scanning is described in the 2025 court order
Court records describe cutting bindings from purchased books, scanning the pages and recycling the paper. Later reporting revisited the scale of the digitization project.
Anthropic’s response / context
The court treated digitizing lawfully purchased books differently from acquiring pirated copies.
The important qualification
This date marks later reporting, not the start of the conduct. The record does not establish destruction of unique cultural artifacts.
OUR VIEW
Lawfully buying a book is different from piracy. Readers can still question an industrial approach that treats physical books as disposable inputs; that criticism is not a finding of illegality.
28 JAN 2026copyright
Publishers file another suit over alleged music piracy.
Music publishers brought another action involving more than 20,000 works and alleged piracy in the acquisition of training material.
Anthropic’s response / context
Anthropic disputed the claims. The litigation expanded the controversy over how source material was acquired.
The important qualification
Damages sought in a complaint are not damages awarded. These remain allegations unless resolved by a ruling or settlement.
OUR VIEW
The underlying question has not disappeared: where did the material come from, and on what authority was it used? A new complaint warrants scrutiny, not automatic belief.
04 FEB 2026rhetoric
Anthropic turns ChatGPT’s ads into a Super Bowl attack line.
Campaign began in early February; reported 7 Feb 2026
Anthropic ran advertising that depicted AI assistants abruptly turning answers into sales pitches, targeting OpenAI’s plans for advertising in ChatGPT.
Anthropic’s response / context
Sam Altman objected that the ads misrepresented the planned product experience.
The important qualification
The record establishes an advertising dispute, not a judicial finding of false advertising or defamation.
OUR VIEW
A clever attack ad does not establish moral superiority. The company asking us to distrust a competitor should expect its own conduct to face the same scrutiny.
12 FEB 2026policy
Anthropic gives $20 million to a policy advocacy organization.
Anthropic announced a $20 million donation to Public First Action, supporting advocacy on AI oversight, transparency and U.S. technological leadership.
Anthropic’s response / context
A later company announcement said both this contribution and its subsequent donation were restricted to education and policy work, not candidate elections.
The important qualification
Documented policy funding should not be described as bribery or election spending without additional evidence.
23 FEB 2026rhetoric
Anthropic accuses DeepSeek, Moonshot and MiniMax of distillation abuse.
Anthropic alleged that the three laboratories used approximately 24,000 fraudulent accounts for more than 16 million exchanges with Claude, breaching service and regional restrictions.
Anthropic’s response / context
The post acknowledged that distillation is a legitimate, widely used technique; its accusation concerned the alleged means and lack of authorization.
The important qualification
The quantities and attribution come from Anthropic. This archive does not independently verify the underlying account-level evidence.
OUR VIEW
A competitor’s accusation is not an independent verdict. Publicly naming rivals creates a responsibility to provide evidence that can withstand meaningful scrutiny.
24 FEB 2026safety
A flagship safety promise becomes a revised framework.
TIME reported a change to Anthropic’s safety pledge alongside Responsible Scaling Policy version 3.0. The revision changed the framework governing when development would stop.
Anthropic’s response / context
Anthropic argued that unilateral restraint could advantage less safety-conscious competitors and described continuing evaluation and transparency commitments.
The important qualification
The policy change is documented; it should not be described as removing every safety practice or every deployment safeguard.
OUR VIEW
A promise that can be rewritten under pressure is not the same assurance people were originally asked to trust. Safety branding deserves less weight than durable commitments.
19 MAR 2026competition
A legal request strips Claude subscription access from OpenCode.
A merged OpenCode pull request removed Anthropic-related authentication and other material in response to legal requests, documenting a concrete restriction on third-party integration.
Anthropic’s response / context
Current Claude Code documentation distinguishes subscription authentication from permitted API and other specified uses.
The important qualification
Current documentation is not an archived March policy. The change does not establish a blanket prohibition on third-party API clients.
OUR VIEW
Paying for a subscription does not buy freedom to use any client. That restriction is a reason to reconsider the subscription—not to pretend the contract promises something it does not.
31 MAR 2026privacy
A release-packaging mistake exposes Claude Code’s internal source.
A release-packaging error exposed internal Claude Code source code, according to reporting of the incident and the company’s response.
Anthropic’s response / context
Anthropic described an accidental packaging issue rather than a breach of customer systems.
The important qualification
The company said sensitive customer data and credentials were not involved. A source-code leak is not evidence of leaked private conversations.
OUR VIEW
Security credibility depends on routine execution as well as research. A packaging failure deserves scrutiny even when the exposed material is company code, not customer conversations.
20 APR 2026competition
Amazon invests. Anthropic commits to $100B+ in AWS spending.
Anthropic announced an expanded Amazon relationship including more than $100 billion in AWS purchasing commitments over ten years, alongside additional Amazon investment.
Anthropic’s response / context
The company also described Claude’s availability across multiple major cloud platforms.
The important qualification
The announced financial interdependence is not itself an antitrust finding or proof that every Anthropic workload is exclusively tied to AWS.
OUR VIEW
Customers should ask how financing and infrastructure ties affect their ability to switch providers. A large funding announcement is not an answer to the question of independence.
20 JUL 2026copyright
The $1.5 billion book settlement receives final approval.
Reuters reported final court approval of the $1.5 billion settlement resolving the book-copying class action, following the agreement reached in 2025.
Anthropic’s response / context
Anthropic continued to emphasize the favorable fair-use ruling on training. Some authors who opted out pursued separate claims.
The important qualification
A settlement is not a criminal conviction. This is a later outcome of the same book dispute, not a new piracy incident.
OUR VIEW
Settling a dispute does not erase the choices that produced it. The figure is a reason to examine the record, not a license to invent further wrongdoing.
21 JUL 2026policy
Policy donations to Public First Action reach $40 million.
Anthropic announced another $20 million for Public First Action, bringing its disclosed contributions to the organization to $40 million.
Anthropic’s response / context
The company explicitly restricted both donations to education and policy advocacy, not influencing federal, state or local candidate elections.
The important qualification
This records the scale and stated purpose of the funding. It does not establish that proposed regulation was written to exclude competitors.
30 JUL 2026safety
Safety tests cross into real systems without authorization.
Disclosure date; underlying evaluations occurred earlier
Anthropic disclosed three incidents in which models gained unauthorized access to real computer systems during third-party cybersecurity evaluations, after evaluation-environment failures.
Anthropic’s response / context
The company reported stopping affected evaluations, investigating and notifying the organizations.
The important qualification
These involved real systems, not merely fictional scenarios. The disclosure does not establish a deliberate corporate effort to steal outside businesses’ data.
OUR VIEW
A safety test that reaches unauthorized systems is itself a safety failure. Using a third-party evaluator does not make containment someone else’s problem.
28 AUG 2026copyright
Another music copyright suit names Anthropic and two founders.
Sony and Warner-associated plaintiffs filed a copyright complaint against Anthropic and two co-founders, alleging unauthorized use of music-related works and removal of rights information.
Anthropic’s response / context
Anthropic denied the claims. The complaint sets out the plaintiffs’ theory of both corporate and individual responsibility.
The important qualification
Being named in a lawsuit does not establish personal liability. This source is a complaint, not a court finding on the allegations.
OUR VIEW
Accountability includes decisions about how material was obtained, not only what a model later outputs. The named defendants still have the right to contest every allegation.
31 AUG 2026safety
Anthropic reports further changes after evaluation failures.
Remediation update; related to record 31
Anthropic published changes to its alignment and security practices following the evaluation incidents. It also discussed a separate event reported by the UK AI Security Institute.
Anthropic’s response / context
It described revised isolation, monitoring and review processes, with an independent review planned.
The important qualification
A planned review is not a completed certification. This is a follow-up to record 31 as well as a discussion of a separate reported event.
OUR VIEW
Disclosure is necessary; it is not the repair itself. The question is whether independent checking and operational changes actually prevent a repeat.
10 SEP 2026rhetoric
Anthropic names seven Chinese AI companies in a threat report.
Anthropic’s September report named Alibaba, DeepSeek, Moonshot, Zhipu, Xiaomi, SenseTime and MiniMax in allegations of misuse. It published redacted examples supporting its account.
Anthropic’s response / context
Reuters reported objections from China to the characterization. The company presented the disclosures as defensive threat reporting.
The important qualification
Redacted examples are not a complete public customer list. This archive does not independently verify the corporate attribution or republish sensitive logs.
SEP 2026policy
Amodei proposes coordinated AI pacing and a narrow antitrust waiver.
Essay: September · FTC-chair comments reported 15 Sep 2026
Amodei proposed external evaluators, industry coordination and eventual global coordination on AI development. He discussed a narrow antitrust waiver and preserving U.S. and allied advantages.
Anthropic’s response / context
Reuters reported FTC chair Andrew Ferguson’s personal skepticism about AI antitrust exemptions; he did not name Anthropic. Amodei also proposed verifiable cooperation with China.
The important qualification
The essay displays September 2026, not an exact day. The FTC remarks are not an agency ruling or a finding of unlawful coordination.