Apple filed suit in the Northern District of California on Friday, alleging trade-secret theft and breach of contract tied to OpenAI's secretive hardware division. The complaint names Tang Tan, OpenAI's chief hardware officer and a 24-year Apple veteran, along with former Apple electrical engineer Chang Liu, who is accused of retaining a company laptop and exploiting a bug to access Apple's confidential file storage after he left for OpenAI.
Apple claims Tan used internal codenames to recruit Apple staff, told candidates to bring "actual parts" to interviews for show-and-tell sessions, and circulated an offboarding playbook to help new hires dodge Apple's exit-security checks. The filing calls OpenAI's device business "rotten to its core" as a result.
OpenAI has denied wrongdoing, saying it has "no interest in other companies' trade secrets." The dispute lands weeks before OpenAI's hardware unveiling and amid a wider legal fallout — Apple's own revamped Siri now runs on Google's Gemini rather than OpenAI's models.
Announced July 13, the agreement centres on ABB's Future Network Model — a plan to replace fragmented, multi-vendor network systems with one standardised digital backbone, service integration layer, and global network operations centre. TCS will design, integrate, and run the ecosystem end to end.
The deal arrives as TCS's AI services revenue climbs past $2.3 billion annualised, even as the company trims headcount domestically — a split increasingly common across Indian IT majors betting on AI-led deal value to outpace traditional staffing models.
The Ministry of Finance had specifically asked about deploying GPT-5.5 for cybersecurity work; the resulting memorandum reflects a broader push to limit dependence on foreign models for functions New Delhi treats as strategic, including work coordinated by CERT-In and India's critical-infrastructure protection body.
The move echoes a parallel story in Washington: both OpenAI and Anthropic recently throttled access to their most capable cyber-oriented models — Anthropic's Mythos 5 and Fable 5 among them — after U.S. officials raised concerns the same vulnerability-discovery capability that helps defenders could be repurposed offensively.
Model routers sit between an application and its model providers, scoring each request for complexity, cost, and latency before deciding whether it needs a frontier model or a cheaper one can do. Published benchmarks on the approach show cost cuts in the 40–85% range while holding onto most of the quality of an all-frontier setup.
Teams are splitting between managed routers such as OpenRouter and Not Diamond, and self-hosted options like LiteLLM, which trade platform fees for more control but hit scaling limits under heavy load. Either way, routing is increasingly framed as production infrastructure rather than an optimisation to bolt on later.