Tech Roundup: DoorDash Tasks, Robinhood Social & More

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

• Meta confirms a critical security incident after an internal rogue AI agent's actions led to the exposure of sensitive data to employees without authorisation.
• Google begins running a "small" experiment replacing news headlines in search results with AI-generated ones, after adding the feature in Google Discover in January 2026.
• OpenAI plans "an autonomous AI research intern" by September and says its "North Star" is to build a fully automated multi-agent research system by 2028.
• WordPress.com says it will now allow AI agents to draft, edit, and publish content on customers' websites, as well as manage comments and update metadata.
• Roblox plans to take a share of revenue from sponsorships in its games and is overhauling advertising policies beginning May 4, 2026.
• Anthropic says it cannot manipulate Claude once the military has deployed it, denying the U.S. Department of Defense accusations that Anthropic could tamper with models during war; launches Claude Code channels, which let users interact with a Claude Code session through Telegram and Discord.
• Amazon is reportedly planning to re-enter the smartphone market more than 10 years after its last attempt with a new device that's centred around Alexa.
• Pinterest's CEO Bill Ready throws his support behind an Australia measure banning social media for younger teens and calls for governments around the world to implement similar bans; says "we need a clear standard: no social media for teens under 16, backed by real enforcement, and accountability for mobile phone operating systems and the apps that run on them." (Pinterest already bars teens under 16 from accessing messaging features and other social features. It also makes teen accounts private by default.)
• A U.S. jury finds Elon Musk intentionally misled Twitter shareholders by disparaging the company in 2022 to buy it for a lower price than his original $44 billion bid.
• Microsoft releases MAI-Image-2, ranked third on Arena AI's text-to-image leaderboard behind only models from Google and OpenAI.
• Google starts consumer beta testing of a dedicated Gemini AI app for the Mac, as it seeks to compete with Mac apps for ChatGPT and Claude.
• Google adds a dedicated Meetings section in Google Chat to help users organise conversation list and keep meeting conversations in one place; also debuts a feature to schedule messages.
• Google updates Messages app for Android with the ability to mention people in group chats using the "@" symbol followed by their name, a new Trash folder to recover deleted messages within 30 days, and share live location with friends and family.
• Opera releases Opera GX for Linux, bringing its gamer-focused browser experience to Debian, Ubuntu, Fedora and openSUSE-based systems.
• DoorDash launches Tasks, a new app that pays delivery couriers in some markets to submit video clips and complete other tasks for training AI models.
• Cursor releases Composer 2 AI model for "frontier-level" coding tasks.
• The U.S. Federal Bureau of Investigation (FBI) acknowledges that it purchases "commercially available information" about American citizens consistent with laws in the country.
• Meta plans to reduce its reliance on third-party vendors for content moderation, in favor of AI tools that it says are better at spotting scams and other tasks; comes as the company pares down virtual reality ambitions as it invests heavily in AI. (The company has also announced that it plans to discontinue support for Horizon Worlds on Quest headsets effective June 15, 2026, but has since clarified that it will be available for the "foreseeable future" for existing games.
• Google announces a new "advanced flow" for Android sideloading that requires a mandatory 24-hour cooling-off period to install apps from unverified developers in an effort to tackle scams and malware.
• Alphabet gives up its controlling stake in its Verily life sciences unit.
• OpenAI plans to unify ChatGPT, Codex, and its browser into a desktop "superapp" to simplify the user experience and focus on engineering and business customers; agrees to acquire Astral, which makes Python tools for developers, to integrate its team into Codex, and says Codex has more than 2 million users, up 3x since January.
• Amazon acquires startup Rivr to test robots for doorstep delivery and to improve safety outcomes for delivery drivers.
• The U.N.-backed World Happiness Report says passively consuming algorithmic social media hurts teens' mental health and heavy use disproportionately affects girls.
• U.K. watchdog Ofcom fines 4chan £520,000 for Online Safety Act (OSA) violations, including £450,000 for not complying with age check requirements, and gives it until April 2, 2026, to implement age assurance.
• The U.K. Competition and Markets Authority (CMA) launches an investigation into Adobe, saying early membership cancellation fees for certain products might be in breach of consumer protection laws.
• Adobe launches Firefly Custom Models in public beta, letting users train AI image generators on their own assets; the custom models are private by default; to pay a $75 million fine and provide $75 million in free services to users to resolve a U.S. government lawsuit accusing it of concealing termination fees.
• Google gives Fitbit's AI health coach the ability to read users' medical records for more personalised health advice, becoming the latest company after OpenAI, Anthropic, Microsoft and Amazon to unveil AI-powered healthcare features.
• Amazon's next-generation smart assistant enters its Early Access program in the U.K., marking Alexa+'s European debut following rollouts in the US, Canada and Mexico; adds 1-hour and 3-hour delivery options in the U.S.
• Google updates Stitch, which lets users turn natural language prompts into UI designs, introducing an "AI-native" design canvas and a reasoning design agent.
• Apple stops vibe coding apps from pushing updates, citing rules on running code.
• The U.K. government withdraws a proposal to let AI companies train on copyrighted works unless creators opt out, following backlash from artists.
• Meta's Facebook launches Creator Fast Track, offering big Instagram, TikTok and YouTube creators guaranteed monthly pay and boosted reach to post on Facebook; launches new tools to help creators detect and report impersonation, and updates guidelines to better define what it considers to be "original content."
• Perplexity releases its Comet browser app for iOS and iPadOS with a built-in AI assistant, four months after launching on Android.
• Cloudflare says it has appealed a €14.2 million fine issued by Italy over the company's refusal to block access to piracy websites on its 1.1.1.1 DNS service, calling the amount "staggering."
• Anthropic releases a survey of 80,508 Claude users' views, hopes, and fears about AI, calling it "the largest and most multilingual qualitative study"; says people seek from AI professional excellence, personal transformation, life management, time freedom and financial independence
• Apple makes an update to Family Sharing in iOS 26.4 (currently in beta) that allows each adult member of the family to use their own payment method for purchases, rather than being forced to share a payment method; adds a new Genius Browse section in the Apple TV app, which gives recommendations for TV shows and movies across multiple suggested categories.
• Robinhood tests a Threads-like social media platform it calls Robinhood Social, allowing users to make posts about their trades and complete trades from their feed.
• Chinese electronics and car manufacturer Xiaomi releases MiMo-V2-Pro, a new 1-trillion parameter foundation model with benchmarks approaching those of U.S. AI giants OpenAI and Anthropic, but at around a seventh or sixth the cost when accessed over proprietary API.
• Google expands Personal Intelligence, which lets Gemini tailor its responses by connecting to Gmail and other Google services, from paid users to all U.S. users.
• OpenAI launches GPT-5.4 mini and nano, aimed at agents, coding and multi-modal workflows, and offering near GPT-5.4-level performance at a much lower cost.
• Mistral announces Mistral Forge to help enterprises build custom models actually trained on their own data using Mistral open-weight models as a starting point; releases Small 4, its first model to unify reasoning, multimodal and coding capabilities of its flagship Magistral, Pixtral and Devstral models
• A U.S. appeals court puts on hold an earlier ruling that had blocked Perplexity from using its agentic shopping tool to shop on Amazon's marketplace.
• Spotify adds a new "Exclusive Mode" in its Windows app that can help maximise audio quality by giving it full control over the device's audio processing; announces Taste Profile editing in beta, the first time it has allowed users to fine-tune the recommendation algorithm, starting with Premium users in New Zealand.
• Baidu integrates OpenClaw with its Xiaodu devices to work as voice-controlled remotes, joining Tencent and Alibaba in racing to capitalise on agentic AI.
• World launches AgentKit, a software development tool that enables sites to verify that a real human is behind the purchasing decisions of AI shopping agents.
• Gamma adds AI image-generation tools in bid to take on Canva and Adobe; debuts Gamma Imagine, which will let users employ text prompts to create brand-specific assets like interactive charts and visualisations and infographics.
• NVIDIA unveils Space-1 Vera Rubin for orbital data centers, saying its GPU delivers up to 25x more AI compute for space-based inferencing compared to the H100; announces NemoClaw, which combines the OpenClaw agent platform with components of NVIDIA's Agent Toolkit to add privacy and security controls.
• Meta-owned Manus introduces My Computer, a Windows and macOS app that enables its AI agent to interact directly with the user's local files, tools and apps.
• xAI, which is already facing multiple investigations around the world over widespread reports that Grok repeatedly created sexualised images of children, is now facing a class-action lawsuit from three teenagers, who allege that photos of them were used by Grok to generate child exploitation material.
• Apple unveils the $549 AirPods Max 2 with an H2 chip, up to 1.5x more effective ANC, studio-quality audio recording and camera remote; acquires MotionVFX, a Polish company that develops plugins, visual effects and motion graphics tools for Final Cut Pro.
• Encyclopedia Britannica and its Merriam-Webster subsidiary sue OpenAI for allegedly misusing their reference materials to train its AI models.
• Chinese e-commerce giant JD.com launches online marketplace Joybuy in the U.K., Germany, France, the Netherlands, Belgium and Luxembourg, as it expands globally to challenge Amazon.
• ByteDance has reportedly suspended the global launch of Seedance 2.0, following cease and desist letters from Netflix, Disney, Paramount, and WBD last month.
• Meta says Instagram will no longer support end-to-end encrypted messages starting May 8, 2026, saying "very few people" were using E2EE in their DMs. (It's worth noting that the feature was never enabled by default and was not widely available.)
• Digg announces a "hard reset" and shuts down operations two months after it was relaunched, citing the scale of AI bot spam.
• The European Commission says "X has submitted remedies in relation to its blue checkmark" in the E.U. and that it will now "assess the proposed remedies."