An Anthropic mathematician says Claude helped him prove math's most famous unsolved problem
Levent Alpöge, a mathematician at Anthropic, the company that makes Claude, posted a short note on his website Friday saying he believes that, working with Claude, he has found a proof of the Riemann hypothesis. It's the most famous unsolved problem in math, a question from 1859 about how prime numbers are spread out, and solving it comes with a $1 million prize from the Clay Mathematics Institute.
There's no paper yet. Alpöge released the proof in Lean, a programming language that lets a computer check every logical step. MIT mathematician Andrew Sutherland ran it through Lean himself and told Scientific American it was a historic moment.
Lean checks every step, but it can only check the problem as it was typed in, so experts need to confirm the code states the Riemann hypothesis correctly. Anthropic hasn't announced the result.
Later on Friday, Scientific American reported that independent mathematicians checking the work are close to finishing and say the proof is almost certainly true. If they confirm it, an AI will have helped settle a problem that has stood for 167 years.
Hackers used a free AI tool to knock out European Commission staff systems for more than 10 days
Hackers used a freely available AI tool in an attack on the European Commission, which runs the European Union's day-to-day business. The Commission confirmed the attack Thursday after Politico first reported it, and a spokesperson called it an "AI-generated cyber campaign."
For more than 10 days, staff couldn't fully use the systems they rely on for payslips, timesheets, time off and medical reimbursements, The Register reported. Salaries still arrived on time, but the payslip and leave system went down again on Wednesday. The Commission says it has found no evidence of a breach affecting staff broadly and no ransom demand, and its investigation is still open.
According to Politico, the EU's cybersecurity team spotted the attackers using ARTEX, an open-source tool that uses commercial AI models to find and exploit security holes automatically. The FBI warned last week that in some ARTEX attacks, hackers went from picking a target to deep access in under three hours.
ARTEX's developer pulled the code from GitHub in October, but copies were already circulating, so anyone who wants it can still get one.
Google's Gemini can now click through apps on your Mac for you
Google on Thursday gave its Gemini app for Mac the ability to use your computer for you. You describe a task, and Gemini handles the clicking, typing and form-filling across your files, Mac apps and websites. Google's examples include pulling weights and addresses out of PDFs to fill in shipping quotes, and turning a folder of clothing photos into listings on resale sites.
It works in a separate background window with a live preview in the chat, so your own screen and mouse stay free while it runs. You choose which apps it can use, you can pause it at any time, and it asks you to approve key actions before it takes them.
TechCrunch reports that it needs a paid Google AI Plus, Pro or Ultra plan and is starting in English in the US. Google hasn't said when it will reach other countries, and it's arriving after OpenAI, Anthropic and Meta already put similar features in their desktop apps.


