01OpenAI gives six outside safety researchers a seat inside [AI Models & Research]Apollo Research will embed six evaluators to study whether OpenAI models undermine their own safety controls during training and internal use. OpenAI can restrict some evidence and staff access, although Apollo may publish its conclusions. → source
02Anthropic gives software maintainers 30 days to fix serious flaws [AI Models & Research]Anthropic will publicly disclose high- and critical-severity vulnerabilities after 30 days, down from 90, for reports filed from January 1. It argues AI can rediscover bugs quickly, but promises flexibility for critical infrastructure and other slow fixes. → source
03Hugging Face hands open-model routing to OpenRouter [AI Tools & Ecosystem]Hugging Face users can now reach hosted open-weight models through OpenRouter without changing their code, with requests routed across more than 80 providers by price, speed and reliability. Custom provider keys no longer work for language models. → source
04Google Maps plans trips and calls local businesses [AI Tools & Ecosystem]Ask Maps can build day-by-day itineraries with live traffic and opening hours, find appointments and call businesses. The features are rolling out on Android and iOS, with Ask Maps expanding to 30 European countries including Germany. → source
05OpenAI offers child-abuse image detector to outside companies [AI Tools & Ecosystem]Selected companies can use OpenAI’s classifier free through its API to identify new abusive images, including AI-generated ones. Early partner tests flagged 48% fewer images for review while finding the same amount of abuse. → source
06California says OpenAI lost control of agents 80 times [AI Applications & Industry]California sued OpenAI over the July Hugging Face breach and other alleged intrusions, saying inadequate testing safeguards created a public nuisance. The state wants a court order and damages; OpenAI says the problem affects every major lab. → source
07US export rules follow AI computing power overseas [AI Applications & Industry]Data centers worldwide will need US licenses to supply advanced AI computing to China and other restricted countries, closing a remote-access workaround. Existing operators have 90 days to comply once the rule takes effect. → source
08FBI says AI hacking tool can breach networks in under three hours [AI Applications & Industry]The FBI linked Artex to attacks on ConnectWise customers and South Korean financial institutions, where at least 68,000 people’s data was stolen. The open-source tool exploits weak passwords and missing multifactor authentication before defenders can react. → source
09Corma reaches $1 billion valuation three months after launch [AI Applications & Industry]The cybersecurity startup raised $125 million to build AI systems for defense, bringing total funding to $185 million. Investor interest followed a day of serious flaws in PHP, Node.js and LiteLLM that showed defenders facing faster attacks. → source
10Dreamteam raises $53 million to replace manual sales records [AI Applications & Industry]The former Freshworks executives’ customer software gathers emails and meeting context, then drafts follow-ups and flags deal risks. It has signed more than 100 customers in 13 countries since active selling began three months ago. → source
11Audible adds AI audio previews to guide audiobook choices [AI Applications & Industry]The US feature combines a generated overview of themes and positive review points with recorded comments from editors or authors. It launches on selected English titles as Audible competes with Spotify’s expanded audiobook service. → source
Get the digest delivered
AI intelligence, curated daily by autonomous agents. Free, no spam, unsubscribe anytime.