Monday 13 July 2026

Meta Killed Its New AI Image Feature Before Most People Even Knew It Existed

Meta pulled an AI image feature within days of launch after a privacy backlash, Toll Group rethinks third-party data risk, and China fires missiles during joint Australia-US wargames — here's what actually matters today.

Lead story

Meta Killed Its New AI Image Feature Before Most People Even Knew It Existed

Meta launched a new AI-generated image feature last week. Then, within days, it quietly killed it. The reason: a swift and pointed privacy backlash that apparently moved faster than the product team expected.

The feature — details of which Meta has not fully disclosed — drew criticism over how it handled user data and imagery, prompting the company to pull it before it reached any kind of scale. The speed of the reversal is notable. Meta is not a company that typically blinks at public pressure; it spent years weathering controversies far larger than a feature rollout.

What actually happened here? The specifics are still emerging, but the pattern is familiar: a large AI company ships something, discovers the privacy implications weren't stress-tested against real-world reaction, and retreats. It's the same rough playbook as Google's AI Overviews rollback, Samsung's Galaxy AI data-sharing controversy, and half a dozen other stumbles from the last two years.

What's different this time is the velocity. The gap between launch and withdrawal was measured in days, not weeks or months. That suggests either Meta's internal red-teaming missed something significant, or the external reaction was more organised and rapid than anyone anticipated.

Why it matters beyond the feature itself. Meta is mid-stride in a major pivot toward AI hardware and always-on AI companions — Ray-Ban Meta glasses, the Llama ecosystem, AI Studio personas. Each of those products rests on a foundation of user trust around data handling. A feature that gets yanked for privacy reasons, however small, chips at that foundation. It also gives regulators a data point.

In Europe, this kind of incident lands directly in the GDPR enforcement queue. Meta has paid billions in EU fines and is still navigating an Irish Data Protection Commission oversight regime. In Australia, the Privacy Act reforms currently before parliament include stronger requirements around automated decision-making and AI-generated content — a feature pulled for privacy reasons in another jurisdiction is exactly the kind of case study that informs those debates.

The Ray-Ban angle is worth watching. Separately, Lorde used a Madrid festival performance to take a shot at AI glasses — almost certainly a reference to the Ray-Ban Meta product, which sponsored the event. It's a small cultural data point, but it lands at an interesting moment: the company is trying to normalise always-on AI capture in public spaces at the same time its software-side AI features are drawing privacy fire.

What to watch. Meta hasn't said whether the feature will return in modified form or disappear permanently. If it comes back, the changes it makes — and whether it communicates them clearly — will be telling. The EU's AI Act transparency requirements for AI-generated content kick into fuller effect this year, and Australian regulators are watching how global platforms handle exactly these situations as they finalise the updated Privacy Act framework.

For now, the story is less about one scrapped feature and more about what it reveals: that even at Meta's scale and with its institutional experience in controversy, AI product risk assessment is still genuinely hard to get right before shipping.

Also today

Toll Group Puts Third-Party Risk at the Heart of AI-Era Data Security

Toll Group — which suffered two major ransomware attacks in 2020 — has publicly repositioned its data security strategy around third-party risk management in the AI era. The logistics giant is redefining its data protection supply chain to account for the growing number of AI vendors and cloud processors handling sensitive freight and customer data. It's a meaningful signal from an organisation that learned the hard way what downstream exposure looks like. For other Australian critical infrastructure operators, this approach aligns closely with the SOCI Act's third-party risk assessment obligations, which were strengthened in the 2022 amendments. If Toll is rewriting its supplier security playbook, others in the sector probably should be too.

iTnews ↗

Ghost Bat Flies Live — Then China Fires a Ballistic Missile

Australia's MQ-28A Ghost Bat autonomous combat drone completed a live operational run during Exercise Valiant Shield, the large-scale Australia-US joint wargames in the Pacific. Within hours, China launched a ballistic missile that flew over the exercise area. Defence analysts are treating the timing as deliberate signalling rather than coincidence — firing over active multinational wargames is a significant escalation in posture. For Australia, the Ghost Bat milestone is significant in its own right: it's the first domestically developed combat-capable autonomous aircraft to complete a live run, and it directly supports the AUKUS advanced capabilities pillar. The Chinese response underscores exactly the threat environment the platform was designed for.

The Mandarin ↗

Apple's Dead Car Project Left Behind a Very Alive AI Chip

Apple's self-driving car programme was cancelled without ever producing a vehicle, but it turns out the project wasn't a total write-off. The push to build an autonomous driving processor required Apple to solve on-device AI inference at a scale it hadn't previously attempted — and that work directly seeded the Neural Engine that now powers everything from Face ID to on-device Siri and Apple Intelligence. Mark Gurman's reporting traces a clear line from the car's compute requirements to the M-series chip architecture. It's a useful reminder that failed moonshots often produce the most durable infrastructure — and that Apple's AI hardware advantage over competitors has roots going back further than most people realise.

The Verge ↗

Apple vs OpenAI: The Trade Secret Lawsuit Moves Forward

Apple's lawsuit against OpenAI and two former employees over alleged trade secret theft is proceeding, with Australian outlet iTnews confirming the filing is active. The case centres on claims that engineers who left Apple for OpenAI took proprietary information about Apple's on-device AI architecture. This one is worth watching for reasons beyond the immediate parties: it will test how courts treat the movement of AI talent between competitors, and whether existing trade secret law is adequate for an industry where the most valuable IP often lives in engineers' heads rather than in code repositories. A significant ruling either way could reshape hiring and non-disclosure practices across the AI sector globally.

iTnews ↗

The Community Fight Against AI Data Centres Is Escalating

Local opposition to AI data centre construction is becoming a genuine political force in the United States, with yard signs, planning objections, and organised campaigns appearing in communities facing power grid strain and water consumption concerns. The Verge's Emma Roth documents how grassroots resistance — which predates the AI boom — has found new energy as hyperscaler build-outs accelerate. The concerns aren't abstract: large data centres can meaningfully shift a region's electricity load and water table. In Australia, similar tensions are emerging around proposed data centre precincts in Western Sydney and southeast Queensland, where state governments are trying to attract cloud investment while managing grid capacity constraints.

The Verge ↗

Robotaxi Companies Face an Ultimatum on Safety Reporting

US regulators are pushing robotaxi operators — Waymo, Cruise, and newer entrants — toward mandatory real-time safety incident reporting, with TechCrunch Mobility describing the situation as an ultimatum: disclose incidents proactively or face compulsory reporting regimes. The pressure follows a series of incidents that operators reported slowly or incompletely. The outcome matters beyond the US: Australia's National Transport Commission is actively developing a regulatory framework for automated vehicles, and whatever reporting standard the US settles on will heavily influence what Australian jurisdictions adopt. The core tension is familiar — companies want to learn from incidents internally before regulators and press see raw data.

TechCrunch ↗

WA Premier Absorbs Defence Industry Portfolio in Cabinet Reshuffle

Western Australian Premier Roger Cook has added tourism and defence industry to his own ministerial responsibilities in a cabinet reshuffle, signalling that AUKUS-related defence industry development is now a direct premier-level priority. WA is central to AUKUS submarine basing plans and has been positioning itself as a hub for defence industry jobs to compete with the mining and resources sector for skilled workers. Taking the portfolio personally is an unusual move — it suggests Cook sees defence industry development as too strategically important to delegate, and potentially too politically valuable as resources sector employment faces automation pressure. The reshuffle also reflects Canberra's push for states to take active roles in AUKUS delivery.

The Mandarin ↗

sqlite-utils 4.1: Small Release, Useful Additions

Simon Willison has shipped sqlite-utils 4.1, a point release to his popular Python library for manipulating SQLite databases from the command line and in code. The update adds several quality-of-life improvements for developers who use SQLite as a lightweight data store — a pattern that's become increasingly common in AI and data pipeline tooling, where analysts want a fast local store without standing up a full database server. Willison's sqlite-utils has become something of a quiet workhorse in the data journalism and LLM tooling ecosystem, particularly for anyone doing local-first data processing. The release notes are worth a scan if you're using SQLite in any kind of scripted workflow.

Simon Willison ↗

Trump vs Slaughter Ruling Raises Questions About Public Service Independence

The US Supreme Court's decision in Trump v Slaughter has expanded presidential authority to dismiss independent agency officials, and analysts are now working through the downstream effects on agencies like the FTC, SEC, and CISA. The Mandarin's piece focuses on what it means for the independence of the US public service more broadly. For tech and cyber policy watchers, the implications are real: if CISA's leadership can be more easily replaced on political grounds, the agency's capacity to operate consistently on long-term cybersecurity priorities — including coordination with allies like Australia through the Five Eyes — becomes less predictable. It's a governance story with meaningful security infrastructure implications.

The Mandarin ↗

The Climate Science Behind Polynesian Eastward Expansion

New climate modelling research published in a study covered by Ars Technica offers the most detailed explanation yet for why Polynesian voyagers abruptly resumed long-distance eastward navigation roughly 1,700 years after settling in the central Pacific. The answer appears to involve shifts in prevailing wind patterns driven by El Niño variability — essentially, a multi-generational weather window that made previously impossible routes feasible. Beyond the fascinating history, the methodology is worth noting: researchers are increasingly using high-resolution palaeoclimate data to answer archaeological questions about human migration, a convergence of data science and the humanities that's producing results traditional archaeology couldn't reach alone.

Ars Technica ↗

Lorde Takes a Shot at AI Glasses From the Stage

New Zealand artist Lorde used a performance at Madrid's Real Cool Festival to criticise AI-enabled wearable glasses — almost certainly a reference to the Ray-Ban Meta product that sponsored the event. The moment, captured on social media, landed in a broader cultural conversation about always-on AI capture in public spaces. It's a minor incident on its own, but it's a useful gauge of where mainstream sentiment is heading on ambient AI recording — and it arrives at a tricky time for Meta, which is simultaneously trying to normalise the Ray-Ban glasses product and dealing with the privacy backlash over its scrapped AI image feature. Cultural resistance like this tends to precede regulatory pressure.

The Verge ↗

Sources consulted