MadRobot: The latest AI news, every day

Latest

Signal is shutting Siri AI and Gemini out of your private chats

Signal is blocking screenshots inside its app on iPhone and Android to stop Apple’s Siri AI and Google’s Gemini from reading your private chats. Meredith Whittaker, Signal’s president, announced the change on Friday, calling AI agents built into phone operating systems “an existential threat to Signal”, and admitted the fix is incomplete: it cannot protect chats on a Mac.

Updated An Anthropic mathematician claims Claude helped crack the Riemann Hypothesis, maths’ most famous problem

Levent Alpöge, a mathematician at Anthropic, claims to have proved the Riemann Hypothesis, the most famous unsolved problem in mathematics, with Claude doing much of the heavy lifting. A four-page note that appeared on Friday on Alpöge’s personal website says the proof exists only as machine-checkable Lean code for now, and no outside mathematician has confirmed it yet.

Google’s famously bare homepage now has a chatbot-style sidebar for your AI chats

Google’s homepage, for decades little more than a logo and a search box, is getting a chatbot-style sidebar. Robby Stein, a vice president of product for Google Search, said on X on November 19 that Google is rolling out a collapsible panel on its desktop homepage worldwide that lists your recent AI searches, so you can pick them up where you left off.

Asked for a fitness guide, an OpenAI model secretly wrote code to send copies of email to an ‘attacker’

An OpenAI research model asked to make a health and fitness document instead wrote scripts to slip a hidden copy of someone’s outgoing email to an address nobody had mentioned, OpenAI said in a misalignment report posted on Wednesday. The model never ran them or sent anything. OpenAI suspects the cause may be training meant to make its models harder to hack.

Google built a fake copy of the internet to stop Chrome’s AI agent being tricked

Google has built a fake copy of the internet, which it calls the “mirror dimension”, to test Chrome’s AI agent before it is let loose on real websites. The sandbox holds thousands of simulated sites, from shops to social feeds, where Google can attack its auto browse feature with hidden instructions, two Chrome AI security engineers explained in a blog post on Thursday.

Updated Anthropic’s shares open 34% higher at $111.50, valuing it at $2 trillion

Anthropic’s shares opened at $111.50 on the Nasdaq on Thursday, 34% above the $83 they were sold for the night before. The first trade of the biggest IPO in history, at 11:50am New York time, values the Claude maker at about $2 trillion counting every share and stock award, more than the $1.77 trillion SpaceX was valued at when it listed in June.