AI news, 13 July: Apple sues OpenAI over trade secrets as Musk and Altman spar publicly

Apple sues OpenAI alleging trade-secret theft for hardware plans. Apple filed a federal lawsuit in Northern California claiming OpenAI systematically obtained confidential information about unreleased products and processes through hiring of more than 400 former Apple employees, including coaching candidates to share details or bring actual parts. Defendants include OpenAI, its hardware unit (formerly io Products, involving Jony Ive), hardware chief Tang Tan, and another former engineer. OpenAI has denied interest in Apple’s secrets and remains focused on its own innovation; the companies previously partnered on iPhone ChatGPT integration. The dispute, reported by CNBC, Reuters and others around 10-11 July, escalated over the weekend of 11-12 July when Elon Musk amplified the filing and traded barbs with Sam Altman on X, turning legal action into the industry’s most visible feud. Significance: it highlights intensifying competition over talent and on-device AI hardware just as OpenAI prepares consumer devices, with potential implications for both firms’ strategies and OpenAI’s anticipated IPO.

SK Hynix debuts on Nasdaq in record $26.5 billion foreign listing. South Korean memory-chip leader SK Hynix began trading on Nasdaq on 10 July after raising about $26.5 billion via American depositary shares—the largest U.S. listing ever by a foreign company, surpassing Alibaba’s 2014 record. Shares rose sharply on debut amid seven-times oversubscription. Driven by demand for high-bandwidth memory essential to AI accelerators such as Nvidia’s, the listing gives U.S. investors direct access to a key supplier and funds further capacity. Reuters and other outlets noted it as clear investor confidence in AI infrastructure despite market volatility. Significance: the capital raise underscores that the AI boom remains heavily dependent on physical hardware supply chains, not just models.

OpenAI’s GPT-5.6 Sol claims proof of decades-old math conjecture. OpenAI announced that its GPT-5.6 Sol Ultra model generated a purported complete proof of the Cycle Double Cover Conjecture (a graph-theory problem open for roughly 50 years since the 1970s) in under an hour, coordinating up to 64 parallel sub-agents. The company timed the claim with broader GPT-5.6 family releases. Coverage in Scientific American and other outlets notes independent verification by mathematicians is still pending, and details such as full peer review or citations remain under discussion. Significance: if validated, it would demonstrate AI moving beyond pattern-matching toward producing novel research-level mathematics, raising both excitement for scientific acceleration and questions about verification standards.

OpenAI rolls out GPT-Live full-duplex voice models. OpenAI released GPT-Live (and a mini variant), powering ChatGPT Voice with a full-duplex architecture that listens and speaks simultaneously rather than in awkward turn-taking. Features include natural interruptions, real-time translation, mid-conversation web search, and agent task delegation. Rollout began around 8 July for many users, replacing prior Advanced Voice Mode on web and mobile according to OpenAI and TechCrunch. Significance: natural conversation is a major usability leap that strengthens voice as a primary AI interface, with clear applications in customer service, accessibility, and everyday tools—while heightening competition with rivals embedding AI in phones and assistants.

Google, Microsoft and enterprise giants back rival AI agent protocol. According to a report in The Information highlighted in industry roundups, Google, Microsoft, Salesforce, Snowflake and ServiceNow have aligned on a shared backend protocol for connecting AI agents to enterprise data and tools. It is framed as a counter to Anthropic’s widely adopted Model Context Protocol (MCP), which has gained de-facto status. All major players also participate in open standards efforts under the Linux Foundation. Significance: control of interoperability “plumbing” could determine default enterprise AI stacks for years, much as earlier internet protocols shaped the web; the move signals intensifying business competition over who sets the rules for agentic systems.