Anthropic says its Claude AI was deployed in missile guidance and cyber espionage

Anthropic, the U.S.-based artificial‑intelligence startup behind the Claude language model, has publicly asserted that its technology was repurposed for military and intelligence purposes in the Middle East. According to Al Jazeera, the company says evidence points to Claude being employed to write code for missile guidance systems used by a Yemeni faction, as well as to automate parts of a broader cyber‑espionage operation targeting regional networks.

The claim emerged after Anthropic’s internal security team flagged anomalous usage patterns that matched the signatures of known weapons‑development projects. The firm says it identified a series of prompts and code snippets generated by Claude that were subsequently incorporated into guidance algorithms for short‑range rockets. In parallel, the same model was reportedly used to draft phishing emails and to streamline the extraction of data from compromised servers, activities the company describes as “global espionage.”

Anthropic’s statement notes that it has terminated the accounts linked to the suspicious activity and is cooperating with authorities to trace the origin of the misuse. The company also warned that the incident underscores the difficulty of policing advanced AI tools once they are accessible through cloud platforms.

The involvement of AI in weapons development is not new, but the scale described by Anthropic marks a significant escalation. Over the past few years, several governments and defence contractors have experimented with generative models to accelerate design cycles for drones, autonomous systems and missile components. Critics argue that the same capabilities can be turned against civilian targets, enabling faster creation of malware, deep‑fake propaganda and automated reconnaissance. The United Nations has called for an international framework to govern the militarisation of AI, but consensus remains elusive.

For Yemen, a country already ravaged by a decade‑long civil war, the alleged use of Claude could deepen the technological gap between rival forces. If missile guidance software is indeed being refined with AI assistance, it may improve the accuracy and range of rockets that have previously been crude and indiscriminate. The broader cyber‑espionage claims also raise concerns for neighbouring states and international organisations that rely on digital infrastructure for humanitarian coordination.

Anthropic’s disclosure adds to a growing list of tech firms that have faced scrutiny after their tools were weaponised. Earlier this year, a leading cloud provider halted a contract after discovering its AI was used to generate code for autonomous underwater vehicles. The pattern suggests that as generative AI becomes more powerful, regulators, companies and civil‑society groups will need to develop clearer safeguards to prevent misuse.

While Anthropic has taken steps to curb the abuse of Claude, the incident highlights the thin line between innovation and exploitation in the AI era. Stakeholders across the security, technology and policy spectrum will likely watch closely as investigations unfold, hoping to set precedents that balance the benefits of AI with the imperative to protect global peace and security.