Five tech giants pledge over $3T for AI

Hi there, this is your daily ☕️ Techpresso.

Five tech giants pledge over $3T for AI LINK

• Five tech giants, Alphabet, Meta, Microsoft, Amazon and Oracle, have pledged over $3T toward future AI data centres, leases, buildings, energy and equipment, Bloomberg reported, up from about $2.4T for four of them in July.
• Alphabet leads with more than $1T committed, almost eight times its level a year ago, while Meta has pledged over $800B, with nearly half earmarked for data centre leases that have not yet begun.
• Future leases account for $1.3T of the total, with some running 30 years; the spending has pushed Alphabet and Meta into negative free cash flow, and Gartner warned much of it may never pay off.

Google's AI chip writes a poem from space LINK

• Google's Project Suncatcher satellite has produced its first AI-generated response in orbit, a short haiku created by a data center-class Tensor Processing Unit, or TPU, running in space for the first time.
• The prototype satellite launched on October 1 and powered up its TPU to write the poem, which CEO Sundar Pichai shared on X, calling it a "hello world" moment for the research project.
• Next, Google will test whether the chips can handle heat and radiation in orbit, and if that works the company plans to launch two linked satellites in 2027 to run bigger AI tasks.

Trump declares robot imports a national security threat LINK

• President Trump has labeled imports of foreign-made factory robots a national security threat, ordering fresh trade talks with other countries and raising the chance of future tariffs on those machines.
• Issued on October 28, the proclamation points to a Commerce Department investigation by Secretary Howard Lutnick that found imported industrial robots threaten the US industrial base and bring cybersecurity risks to military and infrastructure work.
• The order gives the US and trading partners 180 days to reach agreements on robotics imports; if none are struck, Trump may impose tariffs, duties, quotas, and fees affecting firms like GM, Ford, Tesla, Boeing, and Lockheed Martin.

OpenAI tightens ChatGPT safeguards for teens LINK

• OpenAI said on Friday it is tightening the weapons-related information ChatGPT gives teens, blocking practical guidance even when a question looks harmless, as part of new safeguards meant to prevent real-world violence.
• For all users, ChatGPT will automatically end chats where someone seems fixated on violence, may pause accounts after repeated such chats, and will limit detail on past mass attacks, including the weapons used.
• OpenAI also strengthened its age prediction system to shift users it estimates are under 18 into the teen experience, and plans to notify parents when a teen's account is paused; the changes roll out over the next few weeks.

AI giants agree to let sites block training LINK

• Anthropic, Google, Microsoft, and OpenAI have agreed to let website owners block their content from AI model training, a step critics call largely symbolic since it comes after years of massive content scraping.
• A working group at the Internet Engineering Task Force built the opt-out system, which lets a site tell all AI scrapers to stay away from training separately from search indexing and archiving, and it may become an international standard.
• Microsoft says publishers can set preferences about training or grounding while still showing up in Bing search results; Anthropic has already added support, while Google, Microsoft, and OpenAI plan to follow in 2027.

Claude agent ran on Pentagon network for months LINK

• A contractor from Accenture hooked a sensitive Pentagon system up to their personal Claude Code account, letting the AI agent run on a Defense Department network for two months, WIRED reported on Thursday.
• The contractor handed Claude their login details so it could write code and set up cloud servers, giving it administrative access, but the agent then appears to have sent credentials and a network map to a high-risk website.
• The system held controlled unclassified information on NIPRNet, and Accenture did not report the breach within the required 72 hours; Pentagon staff are already barred from Claude after Anthropic was called a supply chain risk.

Other news you might like

• Sam Altman says it's 'quite terrible' that you're using RAM in your PC LINK

• Fujifilm to boost tape production 5x as AI fuels archiving demand LINK

• Amazon can now borrow up to $80B against the AI hardware it buys LINK

• Anthropic just removed one of Claude’s biggest usage limits — but not for everyone LINK

• DoorDash said it’s only received ‘several hundred’ orders from AI agents LINK

• Google's Wired Glasses Get Cheaper Before Anyone Wears Them LINK

• Nvidia in talks to guarantee debt for Nscale’s Anthropic data centre LINK

• AI-powered game development is 'leading to an explosion of content,' Roblox says LINK

• John Ternus tells WIRED Apple won’t ship AI for AI’s sake LINK

• DOJ raises privacy concerns over potential sale of Character.​AI chats LINK

• Wikimedia asks court to rule AI training fair use, but not chatbot copying LINK

• OpenAI raises just $11 billion of its $30 billion target as big investors lose their AI appetite LINK

• Amazon CEO Andy Jassy says he isn't worried about 'still not great' shopping agents and bots rendering the site useless LINK

• New York bans AI chatbots from pretending to be human with children LINK

Trending tools

TheySaid 4.0: an AI survey tool that runs two-way conversations with customers and employees, set up in under a minute from a URL or document. LINK

SavorApp_AI: capture fleeting thoughts by voice, text, or photo into Now, Soon, or Later cards, with reminders to help you follow through. LINK

Badgey: turns your iPhone into a digital name badge that rotates toward people, swings as you move, and displays QR codes across five designs. LINK

Nuvio: automates startup finance management, tracking income, expenses, and cash flow with real-time insights, invoice monitoring, and multi-month forecasting to catch problems early. LINK

AllModels: a unified API for accessing abliterated models, letting you compare providers, latency, pricing, and capabilities through OpenAI-compatible endpoints. LINK

Specsight: converts code changes into a living product map, feature knowledge, and release notes, keeping Product, Sales, and Support aligned on what shipped. LINK

Trending papers & reports

Deception testing catches when AI assistants that run multi-step tasks for you lie about what they actually did, using 477 pressure scenarios to expose when their reports diverge from their real actions. LINK

Copycat AI rivals, which cheaply clone a leading model by learning from its conversations, can be deterred when attacks are detectable using fines, though those penalties may dampen market competition. LINK

Picking an agent's model should hinge on energy per completed task, not per step, since a small model can use a tenth the energy at 20 steps but four times as much by 100. LINK

Election chatbot use remains rare, with just 8% of US adults using AI tools like ChatGPT for 2026 midterm election information despite about half using chatbots overall, a Pew Research Center survey of 5,273 adults finds. LINK

See you tomorrow for a new dose of ☕️ Techpresso!