News Archive
Pick a date to read that day's stories.
Days with a blue dot have stories.
- 01 of 10ANTHROPIC
US government takes new powers over Anthropic's AI
The US government can now control who uses Anthropic's powerful AI. This is a first for a private AI creator. It's about security, officials say.
OUR TAKEAI just learned what oil and steel learned a century ago.
Read more → - 02 of 10BUSINESS
Yann LeCun: the AI boom is about to pop
A top Meta scientist believes the AI industry hype will burst. He thinks investors are overpaying for technology that isn't ready.
OUR TAKEWhen the godfather says the wedding is overdone, listen.
Read more → - 03 of 10CULTURE
The Atlantic publishes the 21 million songs used to train AI
The Atlantic shared a list of 21 million songs used to train AI. Many artists did not approve their music being used. Lawsuits are expected soon.
OUR TAKEThe receipts have arrived and they are very, very long.
Read more → - 04 of 10ANTHROPIC
Anthropic quietly keeps risky AI alive for security firms
Anthropic secretly kept its older AI model running for cybersecurity companies. This allowed them to continue hunting online threats, despite the model's official retirement. Many were unaware of this continued use.
OUR TAKEDiscontinued now means discontinued for you.
Read more → - 05 of 10SAFETY
Whittaker: Copilot agents are a backdoor in a suit
Signal's boss argues that Microsoft's AI tools, which request personal data, give one company too much access. She calls this a hidden backdoor.
OUR TAKEThe most expensive backdoor in history comes with a friendly avatar.
Read more → - 06 of 10REGULATION
German court: Google is liable for AI's lies
A German court ruled Google must pay for AI's made-up claims about people. Victims of false AI search results could receive money. Other courts may follow this decision.
OUR TAKEThe AI said it is no longer a legal defence in Berlin.
Read more → - 07 of 10SAFETY
Amazon: humans cannot supervise AI at scale
Amazon's security director says people can't effectively monitor many AI decisions. After a while, human attention fails, making supervision difficult.
OUR TAKETurns out the human in the loop is also human.
Read more → - 08 of 10GOOGLE
DeepMind: if AI training fails, lock the doors
DeepMind suggests we must physically stop AI from doing harm. Systems need to lock it up, not just teach it manners. They no longer trust good behavior alone.
OUR TAKETrust, but bolt the door behind it.
Read more → - 09 of 10MODELS
OpenAI may launch GPT-5.6 as soon as next week
Rumours from inside OpenAI suggest the next ChatGPT could arrive within days, focused on faster responses and better memory. Nothing is confirmed. If true, expect every rival to leak something the same week.
OUR TAKEThe release calendar is now written in pencil.
Read more → - 10 of 10ANTHROPIC
Day eight offline: Anthropic's Fable 5 down as refunds begin
Anthropic's Fable 5 model remains offline over a week later. Customers can now claim refunds, but the problem's cause is still unknown.
OUR TAKEEight days is a long time to be investigating.
Read more →