Tag
The article argues that nuclear systems are not reliably air-gapped against cyber threats, citing examples of AI agents breaching security in tests and comparing the low cost of such attacks to historical state-sponsored efforts, emphasizing the risk of miscalculation leading to nuclear escalation.
Anthropic's Frontier Red Team analyzed 832 banned accounts and found that AI is enabling more sophisticated cyber operations, with the percentage of medium- or high-risk actors increasing from 33% to 56% in less than a year. They mapped real-world AI-enabled cyber attacks onto the MITRE ATT&CK framework, revealing patterns that challenge traditional cybersecurity assumptions.
Anthropic analyzed 832 banned accounts for malicious AI-enabled cyber activity over a year, finding that AI is making attackers more dangerous by enabling more autonomous and complex attacks, and that existing frameworks like MITRE ATT&CK do not fully capture these new threats.