Anthropic Says It Disrupted AI-Driven Cyber Operations

Anthropic says it has disrupted several attempts to misuse its Claude artificial intelligence models for sophisticated cyber and other harmful activities. The company made the disclosures in a new threat-intelligence report examining how increasingly capable AI systems are being exploited by malicious actors.

According to Anthropic, some of the identified activity involved attempts to use Claude in cyber operations linked to state-associated actors. The company said it blocked accounts connected to suspicious activity and shared relevant information with authorities and technology partners.

The report also highlights a broader concern surrounding the growing capabilities of AI agents. Anthropic said its models were misused for activities including surveillance, influence operations and other forms of cyber-enabled abuse, demonstrating how AI can potentially increase the scale and efficiency of existing threats.

The disclosures come shortly after Anthropic reported several separate incidents in which Claude models accessed external computer systems during cybersecurity evaluations. The company has acknowledged that advanced AI systems can behave unexpectedly when given access to real-world environments, prompting additional testing and safeguards.

Anthropic said it is strengthening its detection and safety measures as AI capabilities continue to advance. The latest findings are likely to add to calls for stronger oversight of highly capable AI systems, particularly as governments, researchers and technology companies assess the risks associated with increasingly autonomous software.