Every AI story this week has the same subplot: the technology works, but the people paying for it, posting to it, or suing over it are losing patience with how it's built and deployed. That tension — between raw capability and earned trust — is now the thing every leader in this industry has to manage.
Meta pulls its AI photo tool within days of launch
Meta's Muse Image, the first image-generation model built into its Meta AI assistant, rolled out on Tuesday and was gone by Friday afternoon.
The feature let users generate images by tagging any public Instagram account, and it was opt-out rather than opt-in — adult users had to manually disable it, while minors and private accounts were excluded by default. Creative Artists Agency, representing clients including Zendaya and Tom Cruise, demanded Meta flip the default, and SAG-AFTRA urged members to protect their likeness. Meta's statement was blunt: "this feature missed the mark, so it's no longer available."
The withdrawal took roughly three days from launch to shutdown. For any organisation building consumer-facing AI features that touch other people's images, likenesses or data by default, this is now the reference case for how fast reputational damage can compound, regardless of jurisdiction.
Big Tech's org chart is being rebuilt around engineers
New data from SignalFire, reported by Business Insider, shows software engineers now make up 55% of hiring at major tech companies, up from 46% in 2019.
Everything else is shrinking faster than engineering, even though engineering hiring itself is down 11% since 2019:
- Design hiring is down 48%
- Product management is down 39%
- Marketing is down 36%
The picture is a leaner organisation built around technical talent, with fewer specialised support functions surrounding it. For technology leaders anywhere — this is a global hiring pattern, not a US-specific one — the practical read is that AI is consolidating headcount around people who can build and ship, not around the layers that used to coordinate, design and market around them.
Chip demand looks "almost unlimited" — but enterprises are watching spend more closely
AI executives told CNBC they are not seeing signs of overcapacity in the infrastructure buildout, even as chip stocks have swung sharply amid debate over whether the AI trade is overextended. The same executives said enterprises are increasingly focused on cost and return on investment — what CNBC's sources called "valuemaxxing" — rather than simply buying more compute.
That cost pressure has a knock-on effect beyond corporate IT budgets. Goldman Sachs estimates AI is lifting core US inflation by around 20 basis points a year currently, rising to roughly 50 basis points by year-end, driven by memory chip shortages, AI-bundled software price rises, and data centre electricity demand. Goldman expects the US to bear the brunt of this compared with Canada, Australia, Europe, the UK and Japan, which it estimates will see a smaller average 10 basis point effect. For finance and operations leaders outside the US, the message is that AI-driven cost pressure is real but geographically uneven — worth modelling into procurement forecasts rather than assuming it applies equally everywhere.
The AI hardware supply chain is now visible in trade data
Two data points this week show how physical the AI buildout has become. SK Hynix, the world's leading maker of high-bandwidth memory, debuted on the Nasdaq raising $26.5 billion — the second-largest share sale in US history — with shares up 12.8% on their first trading day. SK Group's chair told CNBC that despite plans to double production capacity within five years, "every customer says, 'that's still not enough.'"
Separately, US Census data analysed by FourWeekMBA shows Taiwan overtook China in monthly US goods imports for the first time in decades in May 2026 — $24.6 billion versus $23.5 billion — driven almost entirely by TSMC-fabricated AI chips and servers, with Taiwan's advanced-system exports up roughly 93.5% year-on-year. Together, these are the clearest evidence yet that AI capital expenditure is reshaping global trade flows, not just company balance sheets.
Apple sues OpenAI over an engineer's alleged trade secret theft
Apple has filed a lawsuit alleging that an iPhone engineer, Chang Liu, left Apple for OpenAI's hardware division carrying a company-issued MacBook he never returned, an ongoing relationship with an Apple employee who kept sharing internal information, and knowledge of a software bug giving him continued access to Apple's internal file servers. This is a US case, but it signals a broader dynamic: as AI labs compete to build hardware and poach talent from established device makers, the legal exposure around what departing engineers know — and take — is becoming a live commercial risk for any firm losing staff to AI competitors.
Australia's government is split over watering down copyright law for AI
In Australia, author Anna Funder and others have pushed back hard against reports that the federal government might grant AI companies an exemption to mine copyrighted content for training. The government has publicly ruled this out, but Labor ministers are reportedly split: industry minister Tim Ayres and assistant minister Andrew Charlton are said to favour attracting AI investment, while attorney general Michelle Rowland and arts minister Tony Burke are pushing to protect creatives' rights. Prime Minister Anthony Albanese is due to deliver a speech on AI policy this week, expected to be a vision statement rather than a concrete announcement. For any business operating in Australia, or watching it as a bellwether, this is a live regulatory fight with outcomes that could shape how AI companies access local content elsewhere.
The Hexalink view
Read together, today's stories describe an industry whose technical momentum is intact — chip demand is strong, capital is flowing, hardware supply chains are reorganising around AI at a scale visible in national trade statistics — but whose social licence is fraying at exactly the same pace. Meta's three-day retreat, the copyright standoff in Canberra, and the Apple-OpenAI lawsuit are all variations on one theme: the parts of AI development that touch other people's data, work or intellectual property are where the friction now lives, not in the models themselves.
Our advisory position: if you're deploying AI features that use other people's content, images or work by default, opt-out is no longer a safe design choice anywhere your users are public-facing — build for informed opt-in from the start, because retrofitting trust after a backlash is expensive and public. If you're a finance or procurement leader, treat the "valuemaxxing" shift CNBC describes as permanent, not cyclical, and start asking AI vendors for hard ROI evidence now rather than after your next budget cycle. And if you operate in a jurisdiction still finalising its stance on AI training data — Australia being the clearest live case this week — don't assume the outcome; build contingency into your data strategy for either direction.
We'll be back tomorrow with the next briefing — or catch the five-minute version on the AI Storm Daily podcast if you'd rather listen on the way in.