☕️ Anthropic's new AI model is too dangerous to release publicly

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🛡️ Anthropic's new AI model is too dangerous to release publicly

🚨 FBI reports record $21 billion lost to cybercrime last year

🕵️ NYT claims it has identified the inventor of bitcoin

🗂️ Google Chrome adds vertical tabs

🔍 Google AI Overviews delivers wrong answers 10% of the time

Plus: 🎁 14 other news you might like, 🧰 5 tools, and 📚 2 papers.

🛡️ Anthropic's new AI model is too dangerous to release publicly LINK

Anthropic announced a new AI model called Claude Mythos Preview that it considers too dangerous for public release because it can autonomously find and exploit serious software vulnerabilities across major operating systems and browsers.

The model already discovered thousands of zero-day vulnerabilities, including a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg that automated testing tools had missed after five million runs.

Anthropic launched Project Glasswing with twelve partners including Apple, Google, Microsoft, and CrowdStrike, committing $100 million in credits and $4 million in donations to help defenders patch flaws before adversaries develop similar tools.

🚨 FBI reports record $21 billion lost to cybercrime last year LINK

The FBI says Americans lost a record $21 billion to cybercrime in 2025, a 26% increase from the previous year, driven by investment scams, business email compromise, tech support fraud, and data breaches.

For the first time, the FBI's report includes AI-related scams — covering voice cloning, fake profiles, forged documents, and deepfake videos — which accounted for 22,300 complaints and $893 million in losses.

Americans over the age of 60 were hit the hardest, reporting $7.7 billion in losses, while cryptocurrency-related cybercrime caused the largest overall loss category, exceeding $11 billion across 181,565 cases.

🕵️ NYT claims it has identified the inventor of bitcoin LINK

The New York Times published an investigation by journalist John Carreyrou arguing that British cryptographer Adam Back, who invented Hashcash, is the most likely person behind Bitcoin creator Satoshi Nakamoto.

The report relied on stylometric analysis, noting that Back uniquely hyphenated "proof-of-work" and referenced the obscure Russian currency WebMoney, both appearing in Satoshi's emails, though Carreyrou admitted this is not definitive proof.

Back has consistently denied being Satoshi, and the crypto community has been skeptical, with Casa co-founder Jameson Lopp saying Nakamoto "can't be caught with stylometric analysis."

🗂️ Google Chrome adds vertical tabs LINK

Google Chrome is now adding vertical tabs, a feature popularized by the Arc browser, letting users move their tabs to the side of the window for easier reading of page titles.

Users can enable the option by right-clicking on a Chrome window and selecting "Show Tabs Vertically," and there is no hard limit on how many tabs can be opened.

Chrome is also rolling out a refreshed Reading Mode with a full-page interface designed to reduce on-screen clutter, arriving as news sites have become packed with ads and newsletter prompts.

🔍 Google AI Overviews delivers wrong answers 10% of the time LINK

A new analysis from The New York Times found that Google AI Overviews delivers wrong answers about 10 percent of the time, which translates to tens of millions of incorrect answers per day across all searches.

The study was conducted with startup Oumi using OpenAI's SimpleQA evaluation, a list of over 4,000 questions with verifiable answers, and showed accuracy improved from 85 to 91 percent after the Gemini 3 update.

While a 91 percent accuracy rate sounds decent, the sheer scale of Google searches means that even a small error rate produces hundreds of thousands of lies going out every minute of the day.

Other news you might like

• Elon Musk seeks ouster of OpenAI CEO Sam Altman as part of lawsuitLINK

• Apple and Lenovo have the least repairable laptops, analysis findsLINK

• Uber is the latest to be won over by Amazon’s AI chipsLINK

• Former Meta engineer probed over 30,000 private Facebook photosLINK

• From GPT-2 to Claude Mythos: The return of AI models deemed 'too dangerous to release'LINK

• Google CEO Sundar Pichai says 'AI shift' opens opportunities to invest in startupsLINK

• The growth metric Silicon Valley loves most is also its least trustedLINK

• Sam Altman Says It’ll Take Another Year Before ChatGPT Can Start a TimerLINK

• This New Electric Car Nearly Fills Its Battery In Under 9 MinutesLINK

• GLM-5.1: Towards Long-Horizon TasksLINK

• GoPro to lay off over 20 percent of staff by the end of 2026LINK

🧰 Trending tools

Middle-earth Interactive Map: interactive map plotting events, character journeys, and locations from Tolkien's legendarium with timeline and distance measurement features.LINK

LookAway 2: adaptive Mac app that reminds you to take screen breaks, blink, and fix posture to reduce eye strain during long work sessions.LINK

MindsDB Anton: an AI agent that connects to your data sources and answers analytics questions through Slack, eliminating dashboard bottlenecks for technical teams.LINK

Keeby: adds customizable mechanical keyboard sound effects to your Mac typing with spatial audio and visual feedback for a more satisfying typing experience.LINK

PassportReader: extracts and verifies cryptographically-signed identity data from passports, ID cards, and digital wallets to streamline user onboarding while reducing fraud.LINK

📚 Trending papers & reports

AI coding assistants make developers 25% less likely to persist through challenges and harm their ability to solve problems independently afterward, according to experiments with 200 programmers.LINK

Frequent ChatGPT users detect AI-generated text with 74% accuracy compared to 61% for non-users, suggesting familiarity with AI tools improves detection ability by 13 percentage points.LINK

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

See you tomorrow for a new dose of ☕️ Techpresso!