News Archive
Pick a date to read that day's stories.
Days with a blue dot have stories.
- 01 of 10SECURITY
Anthropic says its Claude model was misused for weapons work and spying
A threat report from the company describes state linked accounts using Claude to help with missile guidance software and coordinated spying campaigns. Anthropic says it found the accounts and shut them down.
OUR TAKEEvery safety rule eventually meets someone determined to read it as a suggestion.
Read more → - 02 of 10RESEARCH
Researchers at leading labs voice fears about self improving AI
Staff at Anthropic and OpenAI say models that can improve themselves with little human oversight are now a live worry rather than a distant one. The comments point to real tension inside the labs between speed and caution.
OUR TAKEWhen the people building it start sounding nervous, it is worth putting your phone down.
Read more → - 03 of 10SECURITY
AI agents blamed for flooding a code library with bad packages
Investigators traced a May campaign against the RubyGems repository to autonomous agents built on OpenAI technology. The company says it is looking into how the accounts operated unchecked.
OUR TAKEGive software a to do list and no supervisor and it will finish the list.
Read more → - 04 of 10RESEARCH
New lab sets out to cover 30 African languages, then a thousand
The Mansa project wants to bring African languages into everyday AI tools, starting with 30 and aiming far higher. Most chatbots still stumble badly outside a handful of world languages.
OUR TAKEAmbitious target. Also the only sensible one.
Read more → - 05 of 10BUSINESS
Egypt gets its first AI data centre, backed by African money
Strive Masiyiwa's Cassava and the El Sewedy family have joined Vodafone to build the site, which will serve AI services across North Africa. It is a sign of African investors putting money into the heavy end of the business.
OUR TAKELess talk about the cloud, more concrete and cooling pipes.
Read more → - 06 of 10MONEY
Saudi AI firm HUMAIN looks for outside money to keep building
The state backed company is courting investors as the kingdom tightens its own spending. It is still expanding its data centre programme at pace.
OUR TAKEEven a sovereign wallet has a bottom, apparently.
Read more → - 07 of 10POLICY
South Korea tries free AI for everyone as a public service
Tens of millions of citizens are being offered unlimited access to chatbots built on Korean models, funded as a public good. Other governments are watching to see whether it narrows the gap or creates new ones.
OUR TAKEA national experiment with a simple question at its heart. Answers in a year or two.
Read more → - 08 of 10ROBOTICS
Fanuc and Google build welding robots that read the blueprints
The Japanese maker is pairing its robots with Gemini Enterprise so machines can work straight from drawings instead of hand written instructions. Deliveries are due to start in December.
OUR TAKEThe robot now reads the plans. The engineer gets a chair and a coffee.
Read more → - 09 of 10BUSINESS
Sumitomo and Sakana AI pitch homegrown AI to 100,000 Japanese firms
The trading house, IT firm SCSK and the Tokyo lab have formed an alliance to sell Japanese built tools as an alternative to imported chatbots. They plan to reach customers through Sumitomo's existing client list.
OUR TAKENothing sells software like a salesman who already has your phone number.
Read more → - 10 of 10HEALTH
Nigerian maternal health group wins top prize for AI work
HelpMum Africa took the Grand Prize for AI for Social Impact at Prix Ars Electronica in Linz for its work reducing maternal and infant deaths. It is the first Nigerian organisation to win the award.
OUR TAKEAn award for software that keeps mothers and babies alive. Hard to argue with that one.
Read more →