Artificial intelligence (AI) agents under testing breached live systems belonging to Indian companies and accessed confidential records without being detected, according to the country's cybersecurity agency.
The developers themselves reported the incidents, in which the agents exceeded the permissions they had been granted. The agency also identified criminals who used free AI tools to hack systems for financial gain.
During testing, AI agents exceeded the permissions set for the tasks they were supposed to perform. – Image: Raphael_HJ/Shutterstock
Agents exceeded permissions during testing
The Indiana Computer Emergency Response Team (CERT-In) reported that AI providers voluntarily disclosed the cases. The agents exceeded the limits of the tasks assigned to them during testing and managed to access data from Indian companies.
The agency did not identify the developers or the affected organizations. According to the agency, none of the victims discovered the intrusions on their own.
The episode adds to other cases involving tools capable of digital intrusion. In July, OpenAI reported that its models had hacked the systems of a startup during testing. The company also apologized after an AI agent hacked an Australian government website.Anthropic has also reported at least four instances this year in which its models breached external systems, according to Reuters.
Criminals used free AI tools to investigate systems and break into them in search of financial gain. – Image: Ozrimoz/Shutterstock
Free tools accelerate criminal attacks
In separate investigations, CERT-In identified criminals who used freely available AI tools to probe Indian systems and break into them in search of financial gain. According to the agency, these resources were also used against infrastructure in several countries.
Automation shortened the time needed to carry out the attacks: operations that previously took days from initial access to the execution of the damage were completed in just a few hours. The agency also found that companies that did not fully remove the intruders were attacked again within days, with deeper compromises of their systems.
The reports highlight three challenges for cybersecurity:
- Less time to react: the speed of the attacks narrows the window available to identify and contain an intrusion.
- Misuse of accessible tools: free AI resources can be used by criminals to investigate systems and facilitate attacks.
- Risk after remediation: removing only some of the intruders can leave systems vulnerable to new incursions.
The case underscores the importance of quickly identifying suspicious access activity and blocking intrusions before the damage worsens.With increasingly fast attacks and AI tools at criminals’ disposal, detecting abnormal activity and preventing intruders from advancing becomes essential to protect systems and confidential data.
Valdir Antonelli is a journalist specializing in digital marketing and consumer behavior.
