Anthropic Raises Alarm Over Claude AI Use in Missile Projects, Espionage

From cyber-espionage and weapon development to mass surveillance programmes, Anthropic AI asserts that their Claude models have foiled numerous destructive endeavours. In a recent report, the company asserts that it interfered in northern Yemen to thwart a plan to use Claude as missile guidance software for guided rockets and long-range ballistic missiles, under the conventional weapons front.

The operators allegedly replaced human software developers with Claude. Hence, assigning different instances of the model to write missile-guidance and flight-control software, as stated in Anthropic's threat report.

How Anthropic Thwarted the Breach?

Although many requests were denied by Anthropic's internal protections, a few managed to slip through. To keep their operation a secret, the operators hid their end goals and divided up their work among different sessions so that no one prompt could reveal their true intentions. Anthropic asserted that the operators seemed to have carried out a failed test-fire, but the business stated that it had no proof that the gang was able to field a functional weapon. According to the report, the accounts in question were banned by the corporation.

Additionally, in order to lessen the impact of the threats, it collaborated with both public and commercial organisations by exchanging threat intelligence. According to Anthropic, a Russian-affiliated espionage operation known as Midnight Blizzard (or APT29) nearly entirely relied on automated AI workflows to carry out its tasks. These include phishing, setup, and data theft, targeting Ukrainian, European, and diplomatic targets, as well as drone manufacturers.

At the same time, the business claimed to have thwarted an offensive programme that targeted government and corporate networks in the Middle East, Europe, and Southeast Asia. The operation had been conducted by Chinese students from Hunan province, and Claude had served as its engineering and orchestration layer. Anthropic said it banned the linked accounts and increased surveillance in both instances to catch similar behaviour.

US Administration Vs Anthropic

After a heated disagreement over ethical guidelines, Washington's relationship with the AI company is still tense. The Pentagon placed the corporation on a supply chain risk blacklist earlier this year when it still refused to remove protections against the use of its technology in autonomous weapons and domestic spying. In a recent ruling, a judge in California found that the US Department of Defence had acted unlawfully in awarding the designation, which Anthropic disputed.

There has been a heated public dispute and protracted legal dispute, but the Pentagon allegedly used the firm's Claude models in military operations in Venezuela and Iran. The corporation is currently trying to recover from the Pentagon's blacklisting and regain its full status within the US defence industrial base.