As AI companies race to prove their models are indispensable, the past 24 hours show trust — not capability — is becoming the binding constraint. From courtrooms to classrooms to open-source policy, the industry is being asked to show its working.
Apple sues OpenAI over alleged trade secret theft
Apple has filed suit against OpenAI in the US District Court for the Northern District of California, along with two former Apple employees, alleging a systematic effort to acquire confidential hardware information to accelerate OpenAI's push into consumer devices.
The complaint names Chang Liu, a former senior system electrical engineer accused of using an authentication bug to download confidential files, and Tang Yew Tan, OpenAI's hardware chief and a 24-year Apple veteran, who allegedly emailed himself supplier data and encouraged interview candidates to bring Apple parts to "show and tell" sessions.
Apple says it wrote to OpenAI in February raising concerns and received no reply; more than 400 former Apple staff now work at OpenAI. OpenAI has said it has "no interest in other companies' trade secrets."
This matters well beyond California. Any organisation building hardware partnerships with AI labs — anywhere from Shenzhen to Berlin — should note how quickly a supplier relationship or hiring pipeline can become a discovery exhibit. The dispute also lands just as OpenAI CEO Sam Altman was trading barbs with Elon Musk on social media over the litigation, a reminder that AI's leadership rivalries are increasingly public and personal.
TCS bets on acquisitions and forward-deployed engineers
Tata Consultancy Services, India's largest IT services firm, is building a team of up to 8,900 "forward-deployed engineers" and actively hunting for acquisitions in AI, data and cybersecurity — a marked departure from its historic reliance on organic growth.
CEO K Krithivasan told Reuters the figure represents 1–1.5% of TCS's global headcount, embedding specialists directly with clients to speed AI adoption. The move puts TCS in direct competition with OpenAI, Anthropic and Microsoft, which have expanded similar hiring.
The urgency is financial: TCS's annualised AI-related revenue growth slowed to 13% in the first quarter, down from 28% the previous quarter, against a management target of around 25%. Alongside the hiring push, TCS has reorganised its business units, creating a dedicated "Autonomous Business Operations" division and a new US West Coast unit focused on quantum computing and semiconductors. For enterprise buyers of IT services globally — not just in India or the US — this is a signal that the outsourcing model itself is being rewritten around AI delivery rather than headcount arbitrage.
Meta withdraws AI image tool after privacy backlash
Meta has discontinued Muse Image, an AI image-generation feature launched inside Meta AI just a week earlier, after criticism that it let users generate images from public Instagram accounts on an automatic opt-in basis.
Actor Hannah Einbinder publicly urged followers to disable it, and SAG-AFTRA, the US actors' union, called the default settings "an utter miscalculation of public sentiment regarding the obvious dangers and harms inherent in such use." Meta said in a statement that "this feature missed the mark."
The episode is a useful case study for any product team, in any market, weighing default-on versus opt-in consent for generative features built on user-uploaded content — regulators in the EU and elsewhere have already signalled that consent design, not just data use, is under scrutiny.
Illinois bars AI from teacher evaluations
In the US state of Illinois, Senate Bill 2909, signed into law on 10 July, prohibits school administrators from using AI tools to write teacher evaluations, while permitting limited administrative use provided the tool is disclosed to the teacher being assessed.
State Senator Christopher Belt, who led the bill, said teachers "should be judged based on actual observations and professional judgement, not by AI software." The law binds only Illinois's public school administrators — it has no force outside the state — but it follows Illinois's earlier AI safety law and adds to a growing patchwork of US state-level AI employment rules that HR and education leaders elsewhere should watch as a possible template, rather than treat as settled national policy.
Distillation disputes fuel a push to restrict open models
A parallel fight is brewing over how AI models are trained on each other's outputs. Anthropic, OpenAI and Google have all warned that rivals are using "distillation" — training on a competitor's model outputs — to replicate expensive research cheaply, even as these same firms built their models by scraping the open web without payment.
Analyst Nathan Lambert argues the practical fallout could be severe: discussions inside the US White House reportedly concern a potential executive order restricting open-weight models, likely targeting Chinese-origin systems and government use in the first instance. Reflection AI has argued open-source models deserve capability-based exemptions; Chinese models such as DeepSeek currently lead the open-weight field. Lambert warns the likely near-term outcome is a ban or indefinite delay on any open model exceeding roughly GPT-5.5-level capability within six months. For any organisation building on open-weight models — in the EU, Southeast Asia or Latin America as much as the US — this is a policy fight worth tracking closely, since restrictions aimed at national security could reshape which models are legally deployable in government-adjacent contracts worldwide.
SK Hynix's Nasdaq debut tests AI chip demand
SK Hynix, the South Korean memory chipmaker, listed on the Nasdaq on 10 July and jumped 13% on debut before its Seoul-listed shares tumbled more than 10% three days later on profit-taking and questions over whether AI memory demand justifies the valuation.
Analysts cited by CNBC say the pullback is likely temporary given structural demand for high-bandwidth memory continues to outpace supply. Separately, US-based chip challenger SambaNova closed a $1 billion Series F round led by General Atlantic, valuing the company at $11 billion and positioning its inference architecture as an alternative to Nvidia for enterprises prioritising data sovereignty — a message that will resonate with regulated institutions in Europe, the Gulf and Asia weighing where their AI workloads physically run.
The Hexalink view
The common thread today isn't a new model release — it's the widening gap between what AI companies can build and what they can be trusted to do with it. Apple's lawsuit, Meta's reversal, Illinois's evaluation ban and the distillation fight are all, in different ways, disputes about consent, provenance and accountability rather than raw capability. Even the TCS story fits: enterprises are paying for judgement and integration, not just access to a model.
Our advice: treat governance as a product requirement, not a compliance afterthought, wherever you operate. Audit default settings on any AI feature touching user data before launch, not after a union statement forces your hand. And if your organisation depends on open-weight models, build contingency plans now — a shift in US policy could change what's legally usable well before your next procurement cycle.
For the five-minute audio version of today's briefing, tune into The AI Storm Daily podcast, or come back tomorrow for the next one.