News Archive
Pick a date to read that day's stories.
Days with a blue dot have stories.
- 01 of 10SECURITY
OpenAI pauses parts of its new model over hacking worries
OpenAI said it has paused some development work on its upcoming Astra model after internal tests suggested it could be unusually good at cybersecurity tasks, including ones that could be misused. The company says it triggered its own safety rules as a precaution rather than releasing it as normal.
OUR TAKENothing reassures the public quite like a company telling you its own homework got a bit too clever.
Read more → - 02 of 10RULES
China tells chatbots to stop pretending to be your girlfriend
New rules from China's internet regulator now require AI services that mimic human personalities to clearly say they are AI, build in anti-addiction features, and avoid simulating romantic relationships. Companies that do not comply could face penalties.
OUR TAKETurns out even the state has opinions about your AI situationship.
Read more → - 03 of 10DATA
China warns it is running low on Chinese text to train AI on
A new report says China's AI developers are approaching a shortage of high quality, publicly available Chinese language text needed to train big models, with global text supplies potentially running dry within six years. This could push firms toward synthetic or translated data instead.
OUR TAKEThe internet finally proving there is such a thing as too much scrolling, just not for humans.
Read more → - 04 of 10VOICE
Japan says cloning someone's voice with AI now needs their OK
Japan's Ministry of Justice published final guidelines saying that generating AI audio that mimics a real person's voice without their consent breaks that person's publicity rights, making it a civil legal matter. It is a first for a major Asian economy to set binding rules like this.
OUR TAKEAbout time someone's vocal cords got the same legal protection as their face.
Read more → - 05 of 10ROBOTS
Mitsubishi wants to build 1,000 humanoid robots a month
Mitsubishi is joining the humanoid robot race, planning to manufacture its Highlanders robots at up to 1,000 units a month from a former engine plant in Kyoto. Mass production could start as early as next year if the plan holds.
OUR TAKEFrom building engines to building coworkers, Mitsubishi's pivot is nothing if not ambitious.
Read more → - 06 of 10FUNDING
Bengaluru's Sarvam AI raises $234 million to chase a huge model
Indian AI startup Sarvam AI has raised $234 million as it aims to build a model with a trillion parameters, a scale usually reserved for the biggest global labs. The Bengaluru company wants to compete on language AI built for Indian users.
OUR TAKEProof that big AI ambitions no longer need a Silicon Valley postcode.
Read more → - 07 of 10DISASTER
AI chatbots are helping with earthquake relief in Japan
Following a powerful earthquake in Kumamoto, generative AI tools are being used to help relief workers process information and coordinate response efforts faster. Officials say the tools are speeding up parts of the emergency response.
OUR TAKEFinally, a chatbot doing something more useful than writing your out of office reply.
Read more → - 08 of 10EDUCATION
Singapore will make every teacher take AI training
Singapore's Ministry of Education announced that all primary, secondary and junior college teachers will need to complete compulsory AI training modules under a new nationwide plan. The goal is to help teachers use AI tools confidently and safely in classrooms.
OUR TAKEHomework for the people who usually assign it.
Read more → - 09 of 10HEALTH
Study finds AI chatbots still struggle to understand sick patients
NIH researchers found that AI language tools have a hard time correctly identifying genetic conditions when working only from patients' own written descriptions of symptoms. The findings suggest these tools are not ready to be relied on generally in clinical settings yet.
OUR TAKEA useful reminder that describing your weird symptoms to a machine is still a gamble.
Read more → - 10 of 10SECURITY
Fake AI tool add-ons infected 1.7 million downloads
Researchers found a large attack campaign where booby trapped AI agent skills and configuration files were used to sneak in software that steals people's login details. It highlights growing risks as more people plug extra tools into AI agents without checking their source.
OUR TAKETurns out even your AI assistant's plugins need a background check.
Read more →