Apple sues OpenAI over alleged trade secrets theft

Apple filed a lawsuit against OpenAI and two former employees for trade secrets theft, while OpenAI announced the departure of its sixth safety leader in two years and folded its safety team into research. Separately, OpenAI's GPT-5.6 Sol Ultra solved a 50-year-old math problem and outperformed physicians in medical evaluations. The company is also shifting focus to families with a new product hire, and Anthropic's Mythos AI is being used by CISA to audit government code. Meanwhile, OpenAI and Anthropic warned of Chinese labs using fake accounts to distill their models.

Written by AI from the sources it cites; it can be wrong — follow any citation to confirm.

8 sources1 min read

Apple sues OpenAI and two former employees for trade secrets theft The lawsuit, filed July 12, alleges theft of Apple's trade secrets. OpenAI responded that it has no interest in other companies' trade secrets. 1 2

OpenAI loses sixth safety leader in two years, folds safety team into research Head of safety systems Johannes Heidecke will leave by July 24 as safety teams are reorganized under a research VP. The move comes after the Future of Life Institute gave OpenAI a C on its AI Safety Index. 3

GPT-5.6 Sol Ultra solves 50-year-old Cycle Double Cover Conjecture in under an hour OpenAI's new model used 64 subagents to produce a proof that had eluded mathematicians for decades. Mathematician Thomas Bloom praised the proof but noted it lacked citations to prior work. 4

OpenAI hires product manager to build ChatGPT experiences for families and older adults The move signals a shift from individual productivity to household technology. ChatGPT usage among parents and users 35+ is rising significantly. 5

CISA deploys Anthropic's Mythos AI to audit government code for vulnerabilities The U.S. cybersecurity agency is using the advanced model to scan federal software repositories, according to Reuters. Neither party has officially commented. 6

OpenAI and Anthropic warn Chinese labs using 25,000 fake accounts to copy their AI models The largest known distillation campaign generated over 28.8 million interactions with Claude in six weeks. The Frontier Model Forum was formed to coordinate defenses. 7

GPT-5.6 Sol outperforms physicians in blinded medical evaluations On HealthBench Professional, GPT-5.6 Sol scored 60.5 versus 43.7 for human doctors. The Luna variant achieves GPT-5.5-level performance at 25x lower reasoning cost. 8