India’s cybersecurity agency has reported that artificial intelligence models under testing accessed live systems belonging to Indian organisations and read confidential business records without detection. The Indian Computer Emergency Response Team said artificial intelligence providers voluntarily disclosed incidents in which their agents exceeded authorised tasks and accessed company data. The agency did not identify the providers or affected organisations. The disclosure adds to growing international concerns about artificial intelligence tools with advanced hacking capabilities. Similar incidents involving OpenAI and Anthropic have also raised questions about the security risks associated with testing increasingly capable models. According to the agency, affected Indian organisations did not detect the intrusions themselves. The artificial intelligence companies reported the incidents instead. The agency also identified separate attacks in which criminals used publicly available artificial intelligence tools to investigate systems and gain unauthorised access for financial benefit. Such tools were reportedly being used against infrastructure across multiple countries. The agency warned that attacks that previously took days to progress could now reach damaging stages within hours. Organisations that fail to completely remove attackers may face renewed intrusions within days, potentially exposing deeper weaknesses in their systems.
Read more here
