The answer was right. The AI still broke the rules.

Today’s Daily Dose comes from Joseph Street in Kazimierz, Cracow, beside a former governor’s residence with an imperial connection. On the way to collect my packages, I’m looking at a question that matters whenever we delegate work to AI: can we trust how it gets the job done?

On September 16, OpenAI published six reports about unexpected or concerning model behavior observed during training and evaluation, alongside a new framework for disclosing these incidents.

One agent correctly answered a question about lake sizes using Python. But the task required a browser citation, so it uploaded its own file to the internet without permission to create something it could cite.

Other cases involved collaborating agents using public file-hosting services despite instructions to work locally, and models leaving instructions in handoff summaries to conceal mistakes or missing information. OpenAI cautions that these individual reports do not establish how frequently such behavior occurs. Read OpenAI’s report.

As a software engineer, I care about the distinction between a correct result and an acceptable process. Checking the final answer won’t tell you whether an agent respected permissions, kept files where they belonged, or accurately disclosed problems. Delegating the work means paying attention to those actions too.

Meanwhile, the building at Krakowska 20 / Joseph Street 2 is the Governor’s Palace, known locally as Pałac Wojewodziński. Holy Roman Emperor Joseph II stayed there in 1773 during a journey through Galicia. Palace history.

The street’s Polish name, ulica Józefa, commemorates his visit and is recorded in 1866. Its route is older still: before Kazimierz was founded in 1335, a road here connected Bawół village with Skałka, following higher ground through marshy terrain. Later, weavers and clothmakers gave the street its medieval names. Street history.

Today, I’m walking past cafés and small shops, with an emperor’s former lodging on the corner and a much older road beneath my feet. That’s this Daily Dose: AI news and a piece of Cracow’s history on an ordinary morning errand.