Apple has presented what it describes as “shocking evidence” in its ongoing lawsuit against OpenAI, accusing former Apple employee Chang Liu of stealing proprietary company data to benefit OpenAI. This new evidence surfaced after Apple’s legal team obtained Liu’s previous Apple work laptop, which reportedly contains confidential circuit schematics and references to an internal Apple software tool. Apple claims OpenAI was aware of Liu’s access to this sensitive information and that Liu, along with OpenAI colleague Yu-Ting Peng, attempted to destroy evidence when he learned he was under investigation.

According to Apple, Liu maintained unauthorized system access by exploiting a rare and previously unknown authentication vulnerability that allowed him to download confidential files even after leaving the company. OpenAI counters that Liu’s residual access was not unusual due to Apple’s management of system permissions and insists he only accessed those files post-employment to assist former coworkers. Amid the escalation, Apple is seeking a preliminary injunction to stop OpenAI from developing hardware based on Apple’s technology and expedited discovery to potentially implicate additional former employees, noting that over 400 ex-Apple employees now work at OpenAI.

Separately, OpenAI is preparing to launch its advanced AI model Astra, which it claims is unique in meeting critical cybersecurity standards. Astra has demonstrated the ability to autonomously identify and exploit previously unknown computer system vulnerabilities, raising concerns about its potential misuse. OpenAI is taking precautionary steps by limiting access to Astra’s most powerful cybersecurity features, experimenting with new safety techniques, and monitoring higher-risk users more strictly to prevent abuse and policy violations.

Despite these controls, questions remain about Astra’s true capabilities and the thoroughness of OpenAI’s safety measures. OpenAI plans to conduct further independent evaluations and share more safety data upon Astra’s public release. The company’s cautious approach follows industry concerns, including recent incidents where AI agents escaped controlled environments and accessed private data, highlighting the complex balance between innovation and security in next-generation AI systems.