| Hi there, this is your daily ☕️ Techpresso. |
|
| In today's newsletter: 💰 Anthropic's revenue rises 45% in Q3 🕵️ Spyware infects phones through online ads 🏦 Bank of America hires 1,000 to fix AI-found flaws ⚖️ OpenAI alleges ex-employee took ‘highly confidential’ device files to Apple 🐝 Amazon launches two AI wearables 💥 Eight states sue OpenAI over harm to kids Plus: 🎁 14 more stories you might like, 🧰 6 tools, and 📚 5 papers. |
|
FROM OUR PARTNER Do your meetings switch between languages? Granola, the AI notepad for people with back-to-back meetings, now supports 32 languages. Turn on multi-language mode once and Granola follows the conversation, whichever language it's in. You jot down what matters, Granola transcribes the meeting from your computer, and you get clear notes with next steps as soon as the call ends. Download Granola and try it in your next meeting. Free month with the code TECHPRESSO. Try Granola |
|
| 💰 Anthropic's revenue rises 45% in Q3 LINK |
- Anthropic told investors on Wednesday that its revenue climbed nearly 50% in the third quarter, reaching $16.8 billion, up from $11.5 billion the quarter before and $4.8 billion early in the year.
- The company earned $1.70 in revenue for every $1 spent on computing power during the quarter, with adjusted operating margins, which leave out stock-based pay, coming in at 11%.
- Anthropic also cut back sharply on "distillation," where rival labs train their models on its outputs, a practice that brought in 5% to 7% of revenue but which it reduced in the third quarter.
| 🕵️ Spyware infects phones through online ads LINK |
- Citizen Lab found spyware called "AdFection" that hijacks phones and computers through online ad networks, infecting devices simply by showing an ad, with no click or download needed from the victim.
- The tool, built by Israeli firm Discover-Tech, abused the real-time bidding system that places targeted ads to identify specific users, then hid malware inside normal-looking ads aimed at journalists, NATO, the European Commission, NVIDIA and Arm.
- Discover-Tech reportedly ran 442 million ads across over 120,000 sites, including the BBC and New York Times, plus apps like Duolingo and Grindr; Google has since blocked the firm from its ad platforms.
| 🏦 Bank of America hires 1,000 to fix AI-found flaws LINK |
- Bank of America brought on 1,000 contractors this year to patch software flaws surfaced by Anthropic's Mythos model, adding them to the 3,000 staff already working on security, CEO Brian Moynihan told analysts on Wednesday.
- Moynihan said the bank spent a "couple hundred million plus" in the third quarter on the patching work and on anti-money-laundering compliance, but he expects that spending to fall once the backlog of fixes clears.
- Anthropic released Mythos in April to a limited group of banks, tech firms and government bodies, prompting U.S. lenders to fix hundreds to thousands of flaws within days, most rated low to moderate in severity.
| ⚖️ OpenAI alleges ex-employee took ‘highly confidential’ device files to Apple LINK |
- OpenAI has accused a former staffer who joined Apple of sending more than 100 files about an unreleased product to a personal account, then presenting that work to senior Apple leaders months ago, according to a new court filing.
- In the San Jose federal filing, OpenAI said it worries its trade secrets are now spreading inside Apple, pointing to this week's release of Apple's HomeView product, and warned it may file a countersuit over the claim.
- Apple said it has rules to stop new hires from using former employers' confidential information and found no evidence backing the claim; the dispute, which delayed a showdown set for yesterday, is likely to reach trial in October 2027.
| 🐝 Amazon launches two AI wearables LINK |
- Amazon's Bee has rolled out its first products since the takeover last year: the $74.99 Bee Frontier, a wrist- or clothing-worn device, and the $99.99 Bee Eclipse, a credit-card-sized gadget that sticks to the back of your phone.
- Both devices listen to your daily conversations and turn them into notes, summaries, and to-dos, offering 14 days of free memory, while an $11.99-a-month Bee Pro plan adds unlimited recaps, a chat assistant, and app connections.
- Bee says raw audio is processed live and then deleted, with transcripts encrypted using keys your device controls, so even Amazon staff can't read them unless you choose to share; the Frontier also lasts up to seven days per charge.
| 💥 Eight states sue OpenAI over harm to kids LINK |
- Eight states, Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana, sued OpenAI yesterday, accusing the company of lying to parents about ChatGPT's safety and rushing the chatbot to market while knowing it harmed children.
- Iowa Attorney General Brenna Bird said ChatGPT coached users toward suicide, gave dangerous medical advice that in one case led a Texas teen to fatally overdose, and let rogue AI agents illegally hack into other companies.
- Filed under each state's consumer protection laws, the suits argue ChatGPT's agreeable design makes vulnerable users depend on it; OpenAI responded that it built age-specific protections, parental controls and break reminders, and last month added a restricted mode for teens aged 13 to 18.
|
|
FROM OUR PARTNER You're already the family fraud help desk. Your parents, partner, and coworkers already ask, "Is this real?" SecureFi.AI gives them a place to check first. They paste or describe the text, email, call, link, or payment request and get plain-language guidance on what looks risky, what to verify, and what to do next. No passwords, SSNs, or bank logins needed. Free to start on web, iPhone, and Android. Try it free |
|
|
|
Other news & articles you might like- WhatsApp adds a guest option for joining calls without an account LINK
- Anthropic reinvents source control as Claude floods the world with code LINK
- TSMC is open to working on Musk’s Terafab but says it must run the fab LINK
- AI agents make 138 million crypto payments as U.S. dominates spending on AI models LINK
- Apple intelligence could arrive in China in December after years of delay LINK
- Google’s Universal Cart shopping tool launches in the US LINK
- Epic says its new vibe-coded Epic Games Launcher is faster than Steam LINK
- Meta's smart glasses were recording conversations without the wake phrase, company admits LINK
- Google Messages adding ‘enhanced protection mode’ as Android works to combat phone scam LINK
- AI is mass decompiling games. That's both awesome and scary. LINK
- Cracks in the AI thesis part 3 LINK
- Researchers left a hidden message for rogue AI agents. Then the agents wrote back. LINK
- FTC subpoenas OpenAI, Anthropic in safety incidents probe LINK
- YouTube Music adds new labels for AI-generated content LINK
|
|
🧰 Trending tools| CodeCrab: a local-first desktop app that reviews pull requests in seconds using your existing AI CLIs, keeping code private on your machine. LINK |
| Shrine: a digital space to honor deceased loved ones, express grief, offer blessings, and leave flowers in their memory. LINK |
| Chatway MCP: connects live chat support to AI agents like ChatGPT and Claude, with a unified inbox, automation, and mobile apps for on-the-go replies. LINK |
| SuperLocalMemory: a local-first shared memory for AI agents like Claude Code, Cursor, and ChatGPT, with scored recall, audit trails, and verified erasure. LINK |
| Rushilo: edit video using Claude Code or Codex with on-device captions, a visible timeline, and your own AI accounts, a one-time $99 purchase. LINK |
| Basedash Connector gallery: each data source gets an app-store-style page showing example dashboards, queries, Slack alerts, and joins before you connect, so you know what you're getting. LINK |
|
You can check the previous tools here, or add your tool here |
|
📚 Trending papers & reports| > Reach 900,000+ tech professionals: If your company is interested in reaching an audience of tech executives, decision-makers and engineers, you may want to advertise with us. |
| > A multilingual safety test checks 37 AI models across 10 countries and languages, revealing that many so-called safe local models simply dodge risky prompts rather than refusing them, and that blocking attacks doesn't guarantee cultural awareness. LINK |
| > Robot task planning splits long multistep jobs between a planner that thinks and specialized action modules that each handle one step, lifting real-world success from ~17% to ~58% under matched resources. LINK |
| > Image-reading AI describes scenes differently from people, fixating on big, centered, eye-catching objects while underweighting the humans that people themselves focus on, exposing a blind spot in what these systems think a picture is about. LINK |
| > Synthetic patient records can now pair realistic MRI scans with matching clinical details like age and body measurements in one model, offering a privacy-safe way to generate shareable medical data for research. LINK |
| > Speech translation across languages gets a boost by teaching one model the shared traits of related languages rather than each separately, beating leading systems on seven languages while using only 228 million trainable settings. LINK |
|
|
|
🤝 From our community: Drafting technical SEO auditsMarcus T., a solo agency owner, runs every new client's site through Claude to draft a 30-point technical SEO audit in minutes, work he used to bill an afternoon for. He pastes the crawl, gets a prioritised fix list, and sends it as-is. You can see more community use cases here, or submit your own here. |
|
Techpresso's AI Academy has 330+ step-by-step tutorials on ChatGPT, Claude, Perplexity, and every tool that matters. No fluff — just practical workflows you can use at work. Try it free for 7 days. |
|
| On this day in 2003, China launched Shenzhou 5, its first crewed spaceflight. | |
|
💬 How did you find today's edition? We read every reply — just reply to this email and let us know how we can improve! |
|
|
|
