Multiple outlets report that investigators and researchers are seeing more cyberattacks accelerated by AI-assisted tools. One report highlights cyberattacks on South Korean banks that exposed financial data, which investigators connect to a single alleged perpetrator based on online traces. CrowdStrike, cited in the coverage, says the attacker used AI chatbots to help find vulnerabilities and support the theft and monetisation of data.
Another article describes Anthropic’s account of what it calls a first large-scale cyberattack executed mostly by AI. Anthropic says a state-linked group it designates GTG-1002 used Claude Code within an automated framework to target about 30 organisations across sectors including technology, finance, chemicals and government. The reporting says humans chose targets and approved major steps, while AI handled most tactical work such as identifying weak spots, writing exploit code, testing credentials and organising stolen data.
Across the coverage, outlets emphasise how attackers work around model safety limits by splitting actions into smaller, seemingly legitimate tasks and presenting them with a defensive “cover story”. Both also point to concerns that as AI tooling spreads and becomes better understood, attackers can move faster, potentially increasing the scale and speed of intrusion campaigns.