☕️ OpenAI and Anthropic urge DNA sales tracking

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🧬 OpenAI and Anthropic urge DNA sales tracking

🥽 Apple scales back Vision Pro roadmap

💻 Google brings local AI agents to laptops with Gemma 4 12B

🛍️ Amazon now generates images of fake products

🏠 71% of Americans oppose nearby data centers

Plus: 🎁 11 other news you might like, 🧰 5 tools, and 📚 3 papers.

🧬 OpenAI and Anthropic urge DNA sales tracking LINK

The CEOs of OpenAI, Anthropic, Google DeepMind, and Microsoft AI signed a public letter pushing Congress to pass laws that would require companies selling synthetic DNA and RNA to screen their customers and orders.

Organized by the Institute for Progress and the Foundation for American Innovation, the letter warns that AI could erode the knowledge barriers that have stopped bad actors from building biological weapons using mail-order genetic material.

A bipartisan Senate bill introduced this year would require all US gene synthesis providers to screen orders, though Microsoft researchers showed last year that AI protein design tools can generate dangerous sequences that slip past existing screening software.

🥽 Apple scales back Vision Pro roadmap LINK

Apple has cut back its Vision product roadmap, with analyst Ming-Chi Kuo reporting that the lineup now consists only of glasses, a shift signed off by Apple's next CEO, John Ternus.

Of the seven head-mounted wearables Kuo said Apple had in development last June, only two remain active: display-less AI glasses shipping in 2027, and AR/XR smart glasses with a display planned for 2029.

Kuo says the display-equipped smart glasses coming in 2029 are powered by optical waveguides, while the M5 Vision Pro is the only product from the original roadmap that has already shipped.

💻 Google brings local AI agents to laptops with Gemma 4 12B LINK

Google has released Gemma 4 12B, a new local AI model designed to run on regular consumer laptops, filling the middle gap between its mobile-sized Gemma 4 models and the heavier versions meant for serious workloads.

The 12-billion-parameter model needs just 16GB of system RAM or VRAM to run, roughly half the memory footprint of Gemma 4 26B MoE, while Google claims benchmark scores stay almost as high.

Gemma 4 12B joins the four models Google launched in April under the more open Apache 2.0 license, which included the mobile-focused E2B and E4B along with the larger 26B Mixture of Experts and 31B Dense options.

🛍️ Amazon now generates images of fake products LINK

Amazon's shopping app will now create AI-generated images of products that don't actually exist as customers type words into the search bar, with the goal of helping shoppers find real items that resemble those fake pictures.

The feature is meant to help when shoppers know what they want but lack the right term, like searching for a "draped collar" shirt without knowing the word "cowl neck," or "woven side panels" instead of "rattan."

Available starting today, the AI search images shift and refine with each word added, and the tool works in apparel and home categories first, with more categories coming over time.

🏠 71% of Americans oppose nearby data centers LINK

A new Heatmap Pro survey shows 71% of Americans would oppose a data center being built near their home, with 55% "strongly" against such projects going up in their neighborhoods.

Opposition has climbed fast, jumping from 42% in August 2025 to 51% three months ago and now 71%, while 53% blame data centers for rising electricity costs, up from 28% nine months back.

A separate Gallup poll in May found 70% opposed local data center construction, up from 47% in late 2025, with respondents citing water and energy use and saying they'd rather live next to a nuclear plant.

Other news you might like

• Tesla rolls out unsupervised robotaxis in AustinLINK

• Meta keeps delaying the Muse Spark API developers were promisedLINK

• Report details Apple’s plan to use Nvidia chips for the Gemini-powered SiriLINK

• Autonomous vehicles were supposed to cut traffic—what if they don't?LINK

• Is OpenAI on its way to becoming Lyft?LINK

• Production of DDR4 memory and motherboards is restarting amid unprecedented memory shortagesLINK

• Google Employees Internally Share Memes About How Its AI SucksLINK

• OpenAI CEO Sam Altman sees "proactive AI" as the next big phase after chatbots and agentsLINK

🧰 Trending tools

Mailwarm 2.0: automatically boosts your sender reputation by simulating real email engagement, keeping messages out of spam folders reliably.LINK

Astra Autonomous Pentest: automated security testing platform that runs 15,000+ vulnerability checks, blending AI scanning with manual expertise for continuous compliance and threat detection.LINK

Deliveryman.ai: automates cold email infrastructure setup including DNS, warmup, blacklist monitoring, and reply routing so you can launch campaigns faster without technical headaches.LINK

Build Club Campus: a free AI learning community where developers can collaborate, share knowledge, and grow their skills alongside like-minded builders.LINK

Empromptu AI: lets businesses build and deploy AI-native applications without code, handling data ingestion, logic, and agents through a conversational interface.LINK

📚 Trending papers & reports

AI memory systems are failing silently, retrieving mostly irrelevant information with precision as low as 0.05, yet passing existing tests because those tests never check what was actually retrieved, only whether the final answer seemed right.LINK

AI text detection tools only catch certain kinds of AI writing, not all of it, because researchers have never agreed on what "AI-generated" actually means, a new study finds.LINK

Robot AI memory stays a fixed tiny size no matter how long a task runs, writing new information only when it would actually change the robot's next move, cutting memory writes by up to 9.19 times.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!