AI safety and regulation
We track the safeguards and legal frameworks addressing the risks of artificial intelligence systems. This covers government regulations, safety standards, and institutional policy measures.
Briefs
- Amazon SageMaker now flags slowing AI services by default
- LangChain opens a public beta for managing AI agents’ tool access
- J-POWER and partners plan a trial of drones inspecting Japan’s power lines
- Databricks opens its data pipeline troubleshooting tool to public testing
- AWS puts AI agents’ marketplace purchases on the cloud bill
- Waymo pauses Bay Area robotaxi rides as storm floods local streets
- US data centers are more common in neighborhoods of color, study finds
- Superhuman opens Go to assistants built by customers and partners
- Google books a New Glenn launch for its orbital AI experiment
- Google’s Stitch Loop uses live product data to suggest software changes
- Snapchat lets creators build camera effects with coding assistants
- Singapore plans a way for banks to check what AI agents are allowed to do
- NVIDIA takes local AI from laptops to compact RTX Spark desktops
- Qwant opens a European web index to models on OVHcloud
- MyMems will turn players' memories into game characters
- Meta adds Loblaw grocery checkout to Muse in Canada
- More employers in Littler’s European survey tie job cuts to AI
- India’s Humyn Labs raises 47 million dollars to gather robot training data
- Google opens Genie 4 to subscribers for editing worlds as they explore
- FamilyMart to roll out drink-restocking robots across Japan
- Cloudflare opens Clairic model weights and adds PDF extraction
- Canva lets users switch off generative AI
- Palo Alto Networks agrees to buy Applied Compute for 4.5 billion dollars in stock
- Waymo pauses driverless freeway trips across the US
- Vals tests find safety gaps in simulated AI targeting
- UK lawmakers propose powers to halt high-risk AI training
- Suno can now generate video clips timed to songs
- Reson8 opens the weights of its multilingual speech model
- Replit gives teams spending and approval controls for AI agents
- Orange adds protected hosting in France to its enterprise AI
Columns
- How pipeline automation changes the on-call engineer's job
- When an assistant edits the shared plan, whose work changes?
- Who gets the hour that AI saves?
- Which patch moves to the front of the maintainer’s queue?
- Who resolves the workplace grievance AI helped write?
- Fewer searches can cost a coding agent more
- How did GIB-200K test where the right answer came from?
- Give Muse one dinner to arrange first
- Cobalt 200 and the agent work that stays on CPUs
- What does silence authorize in an artist’s name?
- Who receives the return on Olala’s public support?
- Who pays for the time to learn?
- The employee’s share of Tone’s sales pitch
- The dependencies behind Hugging Face’s unchanged API
- The layer that can say no to an agent’s call
- The comparison hidden behind 19 extra repairs
- Where do 52 visual cases fall short of an accuracy estimate?
- Who supplies the information between a model’s steps?
- What does DFE-Net’s localization map locate?
- When the solution arrives in MathArena
- Measuring a clinical model’s decision threshold
- The direction of a correct answer in XYBench
- What performance does human oversight preserve?
- Who disagrees with the grader?
- When a clinician corrects an error, has oversight succeeded?
- The weight of questions in a model test
- Where the alarm sounds
- Who defines a good report in UNIRG-CXR?
- The laboratory question after the fast answer
- Podqi’s shopping test counts fake stores and warnings together