Pentagon row to $1.5bn copyright settlement: 5 controversies surrounding Anthropic
![]()
Anthropic's rapid rise has also brought a series of controversies.
Anthropic has positioned itself as a safety-focused alternative in an increasingly competitive AI industry, building its reputation around responsible AI development and safeguards against potential misuse.Founded by former OpenAI employees, the company has repeatedly emphasised the need to ensure that increasingly powerful AI systems remain under human control.However, its rapid rise has also brought a series of controversies, ranging from copyright infringement allegations and disagreements with the US government over military applications to concerns about the existential risks of advanced AI.
The Pentagon standoff
The most high-profile dispute erupted in February 2026, when Anthropic refused to remove restrictions preventing Claude from being used for mass domestic surveillance and fully autonomous lethal weapons.
The US department of defence wanted broader authority to use the technology for what it called all lawful purposes.President Donald Trump directed federal agencies to stop using Anthropic’s technology, while defence secretary Pete Hegseth designated the company a national security supply-chain risk.Anthropic challenged the move in court. In September, a US appeals court upheld the Pentagon’s designation in a 2-1 decision, allowing the department to remove Claude from its systems and barring its use for Defence Department work.
Anthropic said it disagreed with the appeals court ruling.
$1.5 billion copyright settlement
Anthropic agreed to pay $1.5 billion to settle a class-action lawsuit brought by authors who accused it of downloading pirated books to train its AI models. A US court granted final approval in July 2026. The agreement covers more than 480,000 works allegedly obtained from pirate libraries, including Library Genesis and Pirate Library Mirror.The settlement followed a distinction drawn by a federal judge in 2025, which said that using copyrighted books to train AI models could qualify as fair use, but making and retaining copies of pirated books was not protected by that finding.However, the settlement did not establish that all AI training using copyrighted material is illegal.
Researcher’s ‘end of the world’ warning
In September, Anthropic researcher Jacob Coxon announced his resignation, warning that AI developers were racing towards self-improving superintelligence without adequately addressing the risks. Coxon, who said he had worked on pretraining research at OpenAI and Anthropic, wrote that companies were “racing straight to self-improving superintelligence and gambling with our lives”.He said people building the technology believed it could “kill us all by the end of the decade”. His comments were a warning about future risks, not evidence that existing AI systems have reached that level of capability. His resignation fuelled debate over whether AI development should slow until safeguards improve.
Claude and the Venezuela raid
Anthropic faced questions after reports that the US military had used Claude during the January 3 operation to capture Venezuelan President Nicolás Maduro.The Wall Street Journal reported that the model was used in the mission, but Reuters said it could not independently verify the report. Axios also reported its use during the active operation, while noting that the precise role was unclear.Anthropic said it could not comment on whether Claude had been used in a specific classified operation. The reports drew attention because the company’s restrictions on military use were already at the centre of its dispute with the Pentagon.
The tasks Claude performed during the raid have not been publicly established.
Changes to AI safety commitments
Anthropic’s safety policies have also come under scrutiny. In February 2026, it substantially rewrote its Responsible Scaling Policy, introducing Frontier Safety Roadmaps and regular risk reports. In July, it changed internal access rules for unredacted risk reports, requiring them to be shared with at least 200 employees rather than all staff with regular security clearance.In October, the company retired a roadmap goal for systematic oversight of training data to assess alignment with Claude’s Constitution. Anthropic said existing reviews were already identifying the issues such checks were designed to catch, and that a later sample review found no additional problems. It said it would reconsider which goals were most effective for alignment.
KioskNews shows a cleaned-up reading view extracted from the publisher’s page — the original always lives on their site, not ours.