Weekly Recap

Early in the week the briefings covered the business race. On the 20th we noted that the AI fight had shifted from model quality to who controls the customer and the access channel. On the 23rd we wrote that AI companies were scaling investment and lobbying while user data protections were fixed only after pushback and complaints. On the 21st Meta came under pressure over its AI data on three fronts in a single day.

Later in the week safeguards kept falling behind. On the 22nd agents were taking on more work while the systems that watch them gave way first. On the 24th and 25th OpenAI took its model training offline, yet an outside audit found gaps in its internal monitoring and three safety researchers were fired. Raising the wall and checking it from outside moved in different directions.