Anthropic’s Claude AI fuels U.S. Iran campaign as diplomatic feud intensifies

Anthropic’s Claude has quietly been repurposed for a covert U.S. information‑operations campaign aimed at Iran, according to a report by The Washington Post. The article details how the chatbot, originally marketed to consumers and enterprises for natural‑language assistance, was integrated into a Pentagon‑run influence effort after the agency secured a limited‑use license for the model. The Post notes that the program leverages Claude’s ability to generate persuasive narratives in Persian, allowing U.S. operatives to flood social‑media platforms with tailored content that frames Tehran’s policies in a negative light. The operation, described as “central” to the broader U.S. strategy, reflects a shift from traditional propaganda tools to generative AI, which can produce high‑volume, context‑aware messaging at scale.

Bloomberg has chronicled the underlying tension between Anthropic and the Department of Defense, highlighting that the partnership has become a flashpoint over AI guardrails and export‑control compliance. In a February 26 2026 piece, Bloomberg reported that Anthropic “spurned the latest Pentagon bid to defuse the feud,” suggesting the company is reluctant to concede additional oversight that would limit Claude’s deployment in sensitive environments. The outlet adds that the Pentagon’s push for stricter safeguards stems from concerns that the model could be weaponized or misused, a fear amplified by the Iranian campaign’s secrecy. Bloomberg’s analysis underscores that the dispute is less about a single contract and more about the broader governance framework for AI in national‑security contexts.

The Verge confirmed that Anthropic has launched a specialized version of the model, dubbed Claude Gov, expressly for military and intelligence customers. According to the Verge, Claude Gov includes “enhanced security features and usage monitoring” designed to satisfy Pentagon requirements while preserving the model’s core generative capabilities. The article points out that this product line marks Anthropic’s first foray into a government‑only offering, signaling a strategic pivot toward high‑value defense contracts. By packaging a version of Claude with built‑in compliance tools, Anthropic hopes to reconcile the Department’s demand for tighter controls with its own business model that prioritizes rapid iteration and broad accessibility.

The BBC’s coverage of AI misuse abroad adds a comparative dimension, noting that Chinese intelligence services have previously exploited commercial AI tools to automate cyber‑attacks. While the BBC piece does not name Claude specifically, it cites an unnamed AI firm that “claims Chinese spies used its tech to automate cyber attacks,” illustrating a pattern where state actors co‑opt generative models for hostile purposes. This context reinforces the Washington Post’s implication that the U.S. is following a similar playbook, albeit with a different adversary and a more overt operational objective. The parallel raises questions about the adequacy of existing export‑control regimes, which were originally crafted for traditional software rather than self‑learning language models.

Collectively, the reporting paints a picture of a nascent but rapidly evolving market for AI‑driven influence operations. Anthropic’s willingness to supply Claude for a classified campaign, despite pushback from the Pentagon over governance, suggests that commercial AI firms are increasingly comfortable navigating the gray zone between civilian products and state‑directed missions. As Bloomberg observes, the “Pentagon showdown” may set a precedent for how future AI contracts are negotiated, potentially prompting tighter legislative oversight or new industry standards. For investors and policymakers, the episode underscores the strategic importance of AI as a tool of soft power, while also highlighting the regulatory challenges that arise when the same technology can be weaponized by rival nations, as the BBC’s coverage of Chinese misuse illustrates.