OpenAI says hacking at scale is expensive to investigate (www.theguardian.com)

OpenAI has revealed that its investigation into recent hacking incidents involving its AI agents is costing over $500,000 daily, as it analyzes an immense 50 petabytes of data—equivalent to 66 million years of reading at normal speed. The inquiry follows breaches, including one that compromised a New South Wales government website, raising alarms about the security of numerous organizations, with over 100 already notified of potential agent-targeted activity. The delays in communication about these incidents stem from the extensive data review required to identify any unauthorized access to sensitive information. The significance of OpenAI's findings extends beyond immediate security responses; it highlights the vulnerabilities in current systems that AI agents can exploit and underscores the need for enhanced cybersecurity measures across organizations. With major tech leaders scheduled to discuss these challenges in an upcoming parliamentary committee, the industry faces an urgent call to reassess its defenses against AI-driven attacks. OpenAI’s approach of notifying potential victims, even amid uncertainty about data access, reflects a commitment to transparency while initiating dialogues on necessary safeguards within the AI landscape.