Claude Cowork is not just a chatbot with additional features; it operates as a local agent on employees’ machines. This allows it to access files, execute shell commands, navigate the web using stored cookies, and connect to various enterprise systems. According to Anthropic, the potential impact of an issue with Cowork hinges on its access and permissions, significantly altering the threat model. A prompt injection attack can have varied consequences based on what Claude can read and execute.
Related articles
New North Korean campaign uses fake coding interviews to steal developer credentials
DPRK-aligned hackers hid malware inside SVG flag images to backdoor developer job interview coding tests. Not one antivirus vendor caught it.
Abbott probes two cyber incidents amid extortion claims
Abbott Laboratories is investigating two separate cybersecurity incidents after confirming unauthorized access to internal legacy Exact Sciences systems in its Cancer Diagnostics business, while also investigating a separate claim that attackers breached its LabCentral portal and stole company data.
HollowByte DDoS flaw bloats OpenSSL server memory with 11-byte payload
A vulnerability dubbed HollowByte allows unauthenticated attackers to trigger a denial-of-service (DoS) condition on OpenSSL servers with a malicious payload of just 11 bytes.