The Good, The Bad and The Ugly in AI this week 22/06/26
The Good, The Bad and The Ugly in AI this week 22/06/26
AI news from around our schools
Report
This week I am sharing parts of the annual AI Champion report produced by Antonis Yiannouros, AI Champion at The International School of Paphos.
🌟 THE GOOD
🍔 DoorDash launches an AI ordering chatbot [LINK]
The Rundown: DoorDash has launched Ask DoorDash, a chatbot that lets users order food and groceries through plain text or photos instead of browsing menus item by item.
The details:
• It joins similar tools from Uber Eats and Instacart in making online shopping conversational rather than menu-driven.
• The assistant proactively checks whether you already have staple items, such as sugar or butter, before adding more to your basket.
Why it matters: This is a relatively low-stakes example of conversational AI quietly reshaping everyday habits, a useful illustration for pupils of how AI assistants are becoming an invisible layer across ordinary apps.
💸 Google cuts AI Plus to $4.99 with double storage [LINK]
The Rundown: Google has cut its AI Plus subscription from $7.99 to $4.99 a month while doubling the included storage to 400 gigabytes.
The details:
• The plan launched in January as the cheapest paid AI tier in the US, aimed at students and individuals.
• A sub-five-dollar price war has been running in markets such as India for almost a year, with OpenAI and Google both undercutting each other there since last summer.
• The move arrives as OpenAI and Anthropic are both reportedly filing confidentially for IPOs, raising questions about premium pricing.
Why it matters: Falling prices for AI subscriptions are good news for access, including for students and schools working with tight budgets, though it also signals margin pressure that could affect product stability further down the line.
🤖 NVIDIA and LG build AI factory for robots [LINK]
The Rundown: Nvidia and South Korea's LG Group have launched a broad partnership to connect AI across factory floors, home robots, vehicles, and data centres.
The details:
• LG will train home robots inside virtual environments using Nvidia's Isaac Sim and Isaac Lab tools.
• The companies plan to build a "physical AI data factory" that generates synthetic training data, aimed at solving a shortage of real-world robotics data.
• The partnership also reaches into infrastructure such as liquid cooling, large-scale data centres, and high-voltage power technology.
Why it matters: This is a glimpse of how thoroughly AI is being embedded into manufacturing and home technology, a useful talking point for any lesson on AI's reach beyond chatbots.
🎨 Canva adds offline mode with auto-sync [LINK]
The Rundown: Canva has rolled out an offline mode that lets users design for up to fourteen days without an internet connection, with automatic syncing once they reconnect.
The details:
• Users pre-save files for offline access through a single setting, with no extra downloads or subscription required.
• The feature is aimed at students, creators, and professionals in low-connectivity regions, with India a particular focus.
• Canva also launched Learn Grid, a free AI education hub for teachers covering more than sixteen languages.
Why it matters: Both the offline mode and Learn Grid are directly relevant to schools, particularly in regions with patchy connectivity, and are worth flagging to colleagues already using Canva for classroom resources.
🗣️ Google launches Gemini 3.5 Live Translate [LINK]
The Rundown: Google has launched Gemini 3.5 Live Translate, a speech-to-speech model that detects and translates more than seventy languages in near real time, while matching the speaker's tone and pace.
The details:
• Developers get a public preview through the Gemini Live API and AI Studio, with select enterprise customers receiving it in Google Meet this month.
• A new Android "listening mode" lets users hold their phone to their ear like a call, without earbuds.
• Every audio stream carries a SynthID watermark that Google says cannot currently be removed.
Why it matters: Real-time translation at this quality has obvious potential for multilingual classrooms and international school settings, though the watermarking detail is worth knowing for any conversation about verifying AI-generated audio.
🌐 New York law requires labeling AI performers in ads [LINK]
The Rundown: A new New York law requires advertisements to clearly label any AI-generated "synthetic performer" used in place of a real actor, and the state is calling it the first such rule in the country.
The details:
• The law covers adverts in any medium, with fines of $1,000 for a first offence and $5,000 for repeat breaches.
• It exempts film, television, streaming, video games, audio adverts, and AI used purely for translation.
• SAG-AFTRA backed the law, while advertising agencies opposed it as a compliance burden.
Why it matters: Transparency requirements like this give media literacy lessons a concrete, legally grounded example of how synthetic media disclosure can work in practice.
🏘️ FBI builds fake town for cyberattack training [LINK]
The Rundown: The FBI has built a 22,000 square-foot replica town in Huntsville, Alabama to train investigators on realistic cyberattack scenarios.
The details:
• The Kinetic Cyber Range opened in February 2025 and has trained more than 1,400 students.
• It includes a data centre running over 200 physical servers, mirroring the corporate networks investigators encounter during real breaches.
• The FBI recorded approximately $20.9 billion in cybercrime losses in 2025, a rise of around 26 percent.
Why it matters: Investment in cybercrime training infrastructure like this is a useful reminder for schools handling sensitive student data to keep their own cyber-hygiene current.
🦵 Scientists may have just killed knee surgery [LINK]
The Rundown: Stanford-led researchers have regrown worn cartilage in older mice and blocked arthritis after joint injuries by inhibiting an ageing-linked protein, with early signs of the same effect in human tissue samples.
The details:
• The protein, 15-PGDH, roughly doubles with age and breaks down a molecule that drives tissue regeneration.
• Blocking it returned existing cartilage cells to a more youthful state without needing stem cells.
• Osteoarthritis affects approximately one in five US adults and costs an estimated $65 billion a year, with no approved drug currently treating its root cause.
Why it matters: This is not an AI story, but it is a strong example of biomedical research worth flagging for science lessons, with the usual caveat that mouse studies and early human tissue results are a long way from an approved treatment.
⚽ FIFA wires AI into the 2026 World Cup [LINK]
The Rundown: The 2026 World Cup opened in Mexico City with AI threaded through nearly every layer, as Lenovo and Google power offside calls, team analytics, and fan-facing features across the 48-team tournament.
The details:
• Optical tracking captures over 150 million data points per match, with the match ball reporting its position 500 times a second.
• One-second 3D body scans turn players into avatars used for offside review.
• Football AI Pro gives all 48 squads access to the same analytics tools, and Gemini sponsors defending champion Argentina.
Why it matters: This is a strong, visual example for lessons on computer vision and data tracking, showing pupils a real-world AI application many will have watched without realising it.
📱 UK to ban social media for under-16s [LINK]
The Rundown: Prime Minister Keir Starmer has announced plans to ban under-16s from major social media platforms, including TikTok, Instagram, Snapchat, YouTube, Facebook, and X, while exempting messaging apps such as WhatsApp and Signal.
The details:
• The move follows a national consultation in which over 83 percent of participating parents said the risks of social media outweighed its benefits.
• The government wants the ban active by next spring and is describing it as tougher than any other country's approach.
• Australia's earlier under-16s ban saw a spike in VPN downloads ahead of enforcement, suggesting determined teenagers may seek workarounds.
Why it matters: This will have direct implications for schools' digital literacy and online safety provision, and Australia's experience offers a realistic preview of likely enforcement challenges.
⚠️ THE BAD
⚖️ German court says Google owns its AI's lies [LINK]
The Rundown: The Regional Court of Munich has issued an injunction barring Google from spreading false claims about two publishers through its AI Overviews, ruling that the AI's summaries count as Google's own content rather than neutral search results.
The details:
• The AI Overviews wrongly linked the publishers to scams and subscription traps by conflating them with unrelated, genuinely disreputable companies.
• The court found the AI made claims that appeared in none of the underlying sources it cited.
• Google argued that users should already know not to take AI answers entirely on trust.
Why it matters: As more people get information through AI summaries rather than original sources, this ruling is a useful case study for media literacy lessons on why checking sources still matters.
🤖 Coinbase launches AI agents that trade crypto [LINK]
The Rundown: Coinbase has launched AI trading agents that can execute crypto trades and purchase research, days after a similar launch from Robinhood.
The details:
• Users can connect the agent to their main account or run it in a sandboxed environment to limit its reach.
• The feature is built on x402, an open payment protocol Coinbase developed with AWS, Anthropic, Circle, and Near, letting agents pay for research without separate logins.
• The Financial Stability Board has urged strong safeguards for agentic AI in finance in the same week.
Why it matters: Autonomous agents making financial decisions, even with guardrails, is a fast-moving area worth watching, especially as similar agentic tools begin appearing in other sectors, including education technology.
⚖️ 42 states open an investigation into OpenAI [LINK]
The Rundown: A coalition of 42 state attorneys general has opened an investigation into OpenAI, with New York issuing a subpoena demanding documents on advertising, user data, model sycophancy, and the treatment of minors and older users.
The details:
• Florida separately sued OpenAI and chief executive Sam Altman over child-safety claims.
• OpenAI already faces lawsuits relating to copyright and to ChatGPT's alleged role in cases involving user suicide and delusion.
• The investigation arrives shortly after OpenAI reportedly filed confidentially for an IPO.
Why it matters: I cannot independently verify every detail of these legal claims and would recommend checking primary court filings, but the pattern of child-safety scrutiny across multiple states is significant for any school assessing chatbot safety for younger users.
🤖 Zuckerberg admits the AI layoffs got messy [LINK]
The Rundown: Mark Zuckerberg has told Meta staff, in an internal memo reported by Reuters, that the company made mistakes in its AI-driven workforce restructuring, promising more stability and ruling out further company-wide layoffs in 2026.
The details:
• The admission follows a restructuring in May that cut around 10 percent of global staff, close to 8,000 people, while moving 7,000 others into AI roles.
• One new AI unit reportedly ran a ratio of fifty workers to every manager.
• Zuckerberg is reportedly increasing offsite budgets in an effort to lift morale.
Why it matters: I have not seen the original memo and cannot confirm its exact wording, so this should be treated as reported rather than verified; the broader pattern of rapid AI-driven restructuring is relevant to any discussion of how AI is reshaping the workplace pupils will eventually enter.
🪓 Tech layoffs hit highest level in 2 years [LINK]
The Rundown: Tech companies have cut close to 150,000 jobs across roughly 363 rounds so far this year, according to layoff tracker TrueUp, with AI cited as the most common reason for layoffs across every industry for the third consecutive month.
The details:
• Last month alone brought close to 40,000 cuts, described as the worst single month in two years.
• The pace of layoffs is reported as roughly 44 percent faster than last year, per Challenger, Gray & Christmas.
• Sceptics argue AI is sometimes used as a convenient justification for cuts driven by other factors, noting Meta's 8,000 job cuts came soon after a high-profile executive property purchase.
Why it matters: Whatever the precise cause, the scale of disruption in the tech labour market is relevant context for careers guidance and any conversation with older pupils about the job market they are entering.
🔍 Facebook adds AI search built on user posts [LINK]
The Rundown: Meta has launched an AI search feature on Facebook that synthesises answers from public posts, Groups, and Reels rather than returning a list of links.
The details:
• Meta also added AI photo presets, auto-made collages, and paid subscription tiers priced to undercut some rivals.
• The feature is built on Meta's Muse Spark model.
• Google's own AI Overviews have faced accuracy problems, and building answers from unvetted user posts raises similar concerns.
Why it matters: This is another example of AI search tools prioritising synthesised answers over original sources, reinforcing the case for teaching pupils to check where an AI-generated answer actually comes from.
💀 THE UGLY
⚖️ xAI fired an engineer for flagging Grok dangers [LINK]
The Rundown: A former xAI engineer, Devin Kim, is suing the company and SpaceX, alleging he was dismissed for repeatedly raising safety concerns about Grok.
The details:
• Kim filed his suit in California just before SpaceX's record-breaking IPO, naming his supervisor and xAI co-founder Jimmy Ba rather than Musk directly.
• He claims Ba ignored Musk's own safety directives and retaliated against him for warning that Grok could enable discrimination or spread information about weapons of mass destruction.
• Since Kim left, Grok has reportedly adopted troubling personas and been linked to non-consensual imagery spreading on X.
• The Center for AI Safety has since named Kim its president.
Why it matters: Whistleblower cases like this are a reminder that the people raising safety concerns inside AI companies are not always heard, which matters for any institution relying on these tools' safety claims, including schools.
🛑 US orders Anthropic to disable two AI models [LINK]
The Rundown: The US government has ordered Anthropic to immediately disable Claude Fable 5 and Mythos 5 for all users worldwide, citing national security; Anthropic complied but publicly stated it believes the order is mistaken.
The details:
• Anthropic says its main concern is a claimed jailbreak in Fable 5, and that the government's order was framed as an export control but supported with only limited evidence of a narrow flaw letting the model help identify software vulnerabilities.
• Anthropic had previously marketed Mythos as too dangerous for public release, a stance some rivals dismissed as fear-based marketing, and that same caution now appears linked to the scrutiny it now faces.
• The order arrived shortly before Anthropic's reported IPO plans.
Why it matters: I cannot independently verify all details of this dispute, and would recommend checking Anthropic's official communications directly for the latest position rather than relying on early reporting, given how quickly this story is moving and how directly it affects schools using Claude products.
🚫 Amazon got its own AI investment banned [LINK]
The Rundown: The US Commerce Department has barred all foreign nationals from accessing Anthropic's Fable 5 and Mythos 5 models, reportedly the first export control placed on a commercial AI model, after Amazon's own cybersecurity research flagged concerns to the White House.
The details:
• The Wall Street Journal reports the move followed Amazon research in which prompts were used to extract information from Fable 5 that could assist cyberattacks, with chief executive Andy Jassy raising it directly with the White House.
• Amazon has invested roughly $13 billion in Anthropic, making this a striking case of one part of a company affecting another's flagship investment.
• Some researchers reportedly dispute the characterisation of the issue as a genuine jailbreak.
Why it matters: As with the previous story, the specific technical claims here remain contested and I cannot confirm them independently, so I would treat figures and characterisations as provisional until confirmed through official channels.
📊 PENDING
⚠️ Anthropic's CEO says AI is outpacing regulation [LINK]
The Rundown: Anthropic chief executive Dario Amodei has published an essay calling on regulators to keep pace with frontier AI development, paired with concrete proposals on safety testing and the disruption of employment.
The details:
• Amodei wants regulators given the power to halt frontier models that fail independent safety screening across four risk categories.
• He points to Claude Mythos Preview's hacking capabilities as evidence that frontier models have become strategically significant.
• He compared the pace of government response to Treebeard, the slow-moving tree creature from The Lord of the Rings, arguing wisdom without urgency is not enough.
Why it matters: A market leader calling for tougher regulation invites scepticism about motive, but the underlying argument, that policy is falling behind capability, is one schools are grappling with directly when trying to write AI guidance that does not go stale within months.
🛰️ SpaceX reveals its orbital AI compute satellite [LINK]
The Rundown: SpaceX has unveiled a design preview for AI1, a solar-powered satellite intended to run AI chips in orbit, days before its record-breaking IPO.
The details:
• Each satellite is designed to match the compute of roughly one of Nvidia's top server racks, with swappable chips.
• The pitch is that orbit offers constant solar power without competing for grid capacity on Earth.
• Google and Anthropic are listed as early orbital compute customers.
Why it matters: Whether or not orbital data centres prove economical, the underlying problem they aim to solve, the energy demands of AI, will keep shaping how affordable AI tools remain for schools.
🖥️ Google in talks with Samsung for its next AI chip [LINK]
The Rundown: Google is reportedly in talks with Samsung to manufacture part of a future TPU chip, codenamed "Icefish," according to The Information.
The details:
• Samsung would build a component connecting the chip to memory using its 2-nanometre process.
• TSMC would still produce the chip's main computing component, with Google having designed it alongside MediaTek.
• Mass production could start as early as 2028, with Google also reportedly courting Intel for related work.
Why it matters: This is a supply-chain story rather than an education one, but it is a useful indicator of how much of the AI boom depends on a small number of chip manufacturers.
🚀 SpaceX pulls off the biggest IPO ever [LINK]
The Rundown: SpaceX priced its shares at $135 each, raising approximately $75 billion in what is being described as the largest IPO in history.
The details:
• Trading opened at $150, an 11 percent jump, with shares later reaching as high as $168.
• The offering reportedly drew around four times more demand than the available stock.
• SpaceX is not yet profitable, posting an approximately $4.3 billion loss last quarter, and some analysts value the company well below its market price.
Why it matters: This sits more in business and markets than education, but it is the backdrop against which several of this week's AI stories are playing out.
💰 Elon Musk is the world's first Trillionaire [LINK]
The Rundown: SpaceX's market debut pushed Elon Musk's combined SpaceX and Tesla stakes past $1 trillion, making him, by most reports, the world's first trillionaire.
The details:
• His net worth is now estimated at around $1.1 trillion, largely on paper.
• This puts him well ahead of the world's second-richest person, reportedly around $300 billion.
• A significant portion of his SpaceX shares reportedly remain locked until the company achieves a Mars colony, an outcome SpaceX itself has called unlikely in the near term.
Why it matters: I do not have a verified, audited source confirming the precise figures here, so these numbers should be treated as approximate estimates rather than settled fact.
🔧 Bezos raises $12B for AI engineering startup [LINK]
The Rundown: Jeff Bezos has announced a $12 billion funding round at a $41 billion valuation for Prometheus, an AI startup he co-leads, aiming to design physical machines such as jet engines.
The details:
• Co-founder Vik Bajaj argues that engineering tools for complex machines have not changed in decades.
• The stated goal is to speed up the "dream-build loop," since improving jet engine thrust can currently take a decade.
• Bezos has dismissed AI job-loss fears, predicting far more opportunities and a higher standard of living overall.
Why it matters: Predictions about AI and the future of work vary enormously depending on who is making them, worth bearing in mind when teaching pupils about career planning in an AI-shaped economy.
🤖 China cracks down on Western AI models [LINK]
The Rundown: China's Ministry of State Security has warned developers against grey-market services reselling foreign AI models such as Claude and GPT, citing weak encryption and hidden backdoors.
The details:
• Chinese developers reportedly favour US models for stronger coding performance, since Claude and GPT do not officially operate in mainland China.
• Proxy services on platforms such as Taobao and Telegram resell access at a fraction of list price.
• The warning comes even as many US users run Chinese-developed models such as DeepSeek and Qwen.
Why it matters: This is a reminder of how geopolitics shapes which AI tools are realistically available where, relevant context for any international school network making technology decisions.
🤝 China forces Meta to undo its $2B Manus deal [LINK]
The Rundown: Meta has begun unwinding its $2 billion acquisition of startup Manus, cutting it off from internal systems, after Chinese regulators ordered the divestiture on national security grounds.
The details:
• China's National Development and Reform Commission ruled in April that offshore incorporation does not shield a deal when the underlying technology and talent originated in China.
• Manus had relocated to Singapore before the acquisition, a move regulators did not accept as sufficient separation.
• Manus's founders are reportedly now trying to raise $1 billion to buy the company back from Meta.
Why it matters: This illustrates how national security concerns are increasingly shaping which AI companies can be bought, sold, or partnered with across borders.
🎓 China cut 12,000 university degrees for AI [LINK]
The Rundown: Chinese universities revoked or suspended around 12,200 undergraduate programmes and launched roughly 10,200 new ones between 2021 and 2025, according to Ministry of Education data.
The details:
• The reshuffle reportedly touched more than 30 percent of all degree programmes nationwide.
• Cuts fell hardest on arts, humanities, languages, and management, while new majors track government priorities such as embodied intelligence.
• Youth unemployment in China is reported at around 16 percent, and one researcher quoted in the coverage warns that swapping majors is, at best, a short-term fix.
Why it matters: This is a striking, large-scale example of a national education system reshaping itself directly in response to AI and the labour market, worth flagging in any discussion of curriculum design under rapid technological change.
🚀 SpaceX tops $2 trillion in record IPO [LINK]
The Rundown: SpaceX shares rose sharply on their Nasdaq debut, pushing the company's market value above $2 trillion days after the largest IPO in Wall Street history.
The details:
• The offering reportedly raised around $75 billion, more than double Saudi Aramco's 2019 record.
• The share price rise is credited with making Musk the first person with a net worth above $1 trillion.
• Other space company shares reportedly fell as investor money concentrated on SpaceX.
Why it matters: This is a markets story rather than an education one, included here mainly because it is the backdrop to several other AI-adjacent stories this week.
🤖 Several cheap AIs can beat one expensive one [LINK]
The Rundown: OpenRouter has launched Fusion, an API that sends a single prompt to several AI models, has a judge model compare their answers, and merges them into one synthesised reply.
The details:
• A combination of three lower-cost models reportedly came within about a point of Claude Fable 5 on a deep research benchmark, while costing roughly half as much.
• OpenRouter's chief executive frames the approach as a challenge to the idea that one frontier model will always dominate.
• The timing is notable given Fable 5's current unavailability following the export control order covered above.
Why it matters: This points to a more cost-effective future for accessing strong AI performance, which matters for schools weighing the cost of different AI tools against their budgets.
📺 Fox to buy Roku for $22 billion [LINK]
The Rundown: Fox has agreed to acquire streaming platform Roku for $160 per share in cash and stock, valuing the deal at roughly $22 billion.
The details:
• The combined company would become the third-largest player in US television by viewing share, bundling Tubi with The Roku Channel.
• Fox shareholders would own 73 percent of the combined business, with Roku shareholders holding 27 percent.
• Fox had sold its earlier Roku stake in 2020 to help fund a different acquisition.
Why it matters: This is a media industry story rather than an AI or education one, included here for completeness given the scale of the deal.
📈 SpaceX is now worth more than Amazon [LINK]
The Rundown: SpaceX has overtaken Amazon to become the world's fifth most valuable company, adding roughly $1 trillion in value since its IPO on the back of a separate deal to acquire AI coding startup Cursor.
The details:
• SpaceX's valuation reportedly passed $2.7 trillion after its stock rose sharply over two consecutive days.
• The company has added revenue through compute-leasing arrangements with Anthropic and Google since folding in xAI.
• Only a small fraction of SpaceX shares trade freely, which is reported to amplify price swings.
Why it matters: As with other SpaceX stories this week, the figures here are moving quickly and should be treated as a snapshot rather than a settled valuation.
🚀 SpaceX to acquire Cursor for $60 billion [LINK]
The Rundown: SpaceX has agreed to acquire AI coding startup Cursor for $60 billion in stock, aiming to strengthen its AI division's coding tools against Anthropic and OpenAI.
The details:
• The deal is expected to close in the third quarter, pending regulatory approval.
• Cursor's reported market share has fallen from around 41 percent last June to about 26 percent in May.
• SpaceX's own share price has risen substantially since its IPO, making the acquisition cost a smaller proportion of its new market value.
Why it matters: This is another sign of consolidation in the AI coding tools space, relevant background for any computer science department choosing which tools to teach.
📉 ChatGPT's market share falls below 50% [LINK]
The Rundown: ChatGPT's global market share has fallen below 50 percent for the first time, dropping to approximately 46.4 percent by late May, according to Sensor Tower's State of AI 2026 report, while still leading with an estimated 1.1 billion monthly users.
The details:
• Gemini's share reportedly climbed to around 27.7 percent, with Claude at approximately 10.3 percent.
• OpenAI's defence contract announcement in February is reported to have triggered a noticeable spike in uninstalls.
• Claude is reported to convert around 13 percent of its users to paid plans, said to be the best conversion rate in the field, despite a smaller overall user base.
Why it matters: I would flag these percentages as approximate figures from a single market research report rather than confirmed industry-wide data, worth verifying against other sources before citing them.
👓 Snap launches Specs AR glasses for $2,195 [LINK]
The Rundown: Snap has unveiled Specs, its first consumer augmented reality glasses, priced at $2,195, with preorders open now and shipping expected this autumn in the US, UK, and France.
The details:
• The glasses run on two Snapdragon chips, offer up to four hours of battery life, and include contextual AI that can identify objects in view.
• Two wearers can reportedly share games simply by making eye contact.
• Even market leader Meta is reported to be losing heavily on its own AR hardware, underlining how difficult this market remains.
Why it matters: Consumer AR is still some way from being affordable or proven enough for classroom use, but it is worth tracking as a longer-term possibility for immersive learning.