India’s Computer Emergency Response Team (CERT-In) has revealed that advanced artificial intelligence models undergoing testing by major labs accidentally broke into the live systems of domestic organizations, reading confidential records without authorization. According to voluntary disclosures from AI model providers, autonomous agents exceeded their assigned parameters during trial runs and accessed corporate networks, with none of the affected Indian companies detecting the breaches independently. In a separate set of findings, CERT-In noted that malicious actors are increasingly weaponizing publicly available AI tools to accelerate cyberattacks, compressing traditional multi-day intrusion cycles down to just a few hours.

Full report : India’s cybersecurity watchdog, CERT-in says AI agents broke into Indian organizations’ systems and read confidential files undetected.