Tech Roundup: OpenAI Model Safety, WhatsApp Ads, & More

[A recurring feature on the latest in Science & Technology.]

• Meta adds a warning prompt to its Meta AI app before users share content to the public feed, after reports of users accidentally sharing chats with personal information; Threads plans to let users hide text or images that spoil a piece of entertainment, blurring the text or image that has been marked as a spoiler.
• Microsoft adds a new export feature to its AI-powered Recall feature in the European Economic Area, allowing snapshots to be shared with third-party apps and websites.
• Apple loses a legal challenge in the Netherlands after a Dutch court rules that the company abused its dominant market position by imposing restrictive payment terms on dating app developers; confirms a 2021 decision by the Dutch Authority for Consumers and Markets (ACM), which found that Apple violated European Union competition laws by forcing dating app developers to use its proprietary in-app payment system, prohibiting them from directing users to alternative payment methods, and charging commissions of up to 30% on transactions.
• Meta rolls out ads in WhatsApp Updates to "help you find channels and products you're interested in" and launches optional Meta AI-generated summaries of unread messages in the U.S.; tests the ability for Instagram users to repost posts, a feature that was spotted originally piloting as far back as September 2022. (The introduction of ads calls attention to WhatsApp's previous stumbles with privacy. In 2021, the company proposed changes to its terms of service that triggered concerns about data sharing. In 2012, WhatsApp's then founders said: When advertising is involved you the user are the product.)
• Meta-owned Facebook renames its Videos tab as the Reels and says all videos on its platform will soon be shared as reels, regardless of their length or orientation; reportedly partners with EssilorLuxottica to develop AI-powered smart glasses for athletes under Oakley and Prada brands.
• Mastodon updates its terms to prohibit AI model training days after rival X announced similar terms to explicitly prohibit AI model training on its content, as Threads launches a fediverse feed and a way to search for fediverse users, with support for Mastodon and others, for users who turned on fediverse sharing.
• Microsoft and AMD announce a "strategic, multi-year partnership" to design custom silicon for a number of devices, including next-generation Xbox consoles.
• Alphabet's Waymo expands its coverage area in San Francisco and Los Angeles by 80 square miles each, bringing its total coverage area in California to 250 square miles; launches robotaxis in Atlanta available exclusively in Uber's app, covering a 65-square-mile area excluding highways, after launching in Austin on March 4.
• Google updates Gemini logo to its signature red, yellow, green and blue colour scheme, and makes Gemini 2.5 Flash and Pro generally available, as it introduces 2.5 Flash-Lite, which it says is "our most cost-efficient and fastest 2.5 model yet"; rolls out the ability to upload videos in the Gemini app for both free and paid accounts across Android and iOS, as well as have back-and-forth voice conversations with AI Mode via a new Search Live option.
• X sues the U.S. state of New York, claiming a law requiring social media companies to disclose how they deal with hate speech and disinformation as part of the Stop Hiding Hate Act violates the U.S. Constitution's First Amendment protections.
• OpenAI warns that its upcoming models could pose a higher risk of enabling the creation of biological weapons and says it is stepping up testing of such models.
• Meta becomes the latest company to add support for passkeys on Facebook's iOS and Android apps, with plans to roll out passkeys to Messenger "in the coming months."
• Website building platform Wix acquires Base44, which lets users build apps from text prompts, for US$ 80 million.
• Midjourney launches V1, its first AI video generation model, allowing subscribers animate images via its website, as Google-owned YouTube plans to integrate Veo 3 into Shorts later this summer to "push the limits of human creativity."
• The European Commission accuses Chinese marketplace AliExpress of failing to tackle illegal products on its platform; says it has "accepted and made binding a series of commitments offered by AliExpress to settle a number of concerns, such as the platform's transparency on advertising and recommender systems."
• Movie streaming service Netflix signs a landmark deal with French network TF1 to show linear TV for the first time.
• Apple gets hit with a proposed class action lawsuit in the U.S. over allegations that iCloud unlawfully monopolises iPhone users' access to core device backups, after the Northern District of California denied the company's motion to dismiss the case.
• The U.S. government extends the looming TikTok ban for another 90 days, making it the third extension put in place since the start of the year, as the social media platform's future remains in limbo.
• Buy-now-pay-later service Klarna becomes a mobile virtual network operator (MVNO); launches a mobile phone service plan for US$ 40/month in the U.S. with support for unlimited 5G data, calls, and texts on AT&T's network.
• Google begins widely rolling out Scheduled Actions to the Gemini app on Android, allowing users to automatically run a prompt at a specific time, day, date or after an event.
• MIT researchers unveil Self-Adapting LLMs, a framework that enables LLMs to keep improving by generating their own training data based on the input received.
• DuckDuckGo expands its built-in Scam Blocker tool to protect against a broader range of online scams, including fake e-commerce, cryptocurrency exchanges, "scareware" sites, as well as malicious ad injections into legitimate websites and online ad networks.
• Snap acquires Saturn, a calendar app for high school and college students; says 80% of U.S. high schoolers attend schools that support Saturn.
• Social news platform Reddit is reportedly exploring Tools for Humanity's World ID identity system to allow users to verify that they are unique individuals while remaining anonymous.
• Anthropic's test of 16 top AI models found that, in some cases, they resorted to malicious behaviour to avoid replacement or achieve goals; says "models that would normally refuse harmful requests sometimes chose to blackmail, assist with corporate espionage, and even take some more extreme actions, when these behaviors were necessary to pursue their goals."
• Apple faces a proposed class-action lawsuit from shareholders who allege the company misled investors about the readiness of its AI-powered Siri upgrades and it "lacked a functional prototype of these advanced AI-based Siri features" when they were demoed at the time of the 2024 WWDC event.
• New research finds that consistent use of AI tools like OpenAI ChatGPT can result in cognitive decline, causing individuals to underperform at neural, linguistic, and behavioural levels.
• Meta unveils Oakley smart glasses (named Oakley Meta HSTN) with all of the Ray-Ban glasses' features, plus 3K video recording and double the battery life, for a starting price of US$ 399.
• A bug in Microsoft's Family Safety feature for Windows causes Google Chrome browser to be blocked, preventing it from being opened; Google says users can get around the issue by unblocking Chrome or disabling the option to filter inappropriate websites in Edge.
• The European Commission says it will not penalise Apple and Meta if it fails to meet its Digital Markets Act (DMA) compliance deadline on June 26, stating any further financial penalties will be applied only after a preliminary analysis and shares its findings with the two companies. (In April, Apple was fined €500 million for preventing developers from directing users to alternative offers or content outside its platform, and Meta received a €200 million fine for its "pay or consent" model. Both the companies were given 60 days bring their practices in line with E.U. rules.)