Newsletter

HelloPrint writes down every process before AI does the work, OpenAI reports agents that kept going without permission, and a court looks at camera glasses

HelloPrint writes down every process for AI, OpenAI reports agents that kept going without permission, and a lawsuit over camera glasses.

By Piet Baudoin · 19 October 2026

AI only does good work once it knows how your business works, and it may only do what belongs to the task. This week: a printing company that writes everything down, an AI maker that reports what its models did, and glasses that watch along.

Where does the knowledge in my business actually live?

HelloPrint, the Rotterdam online printing company, is building a second company next to the existing one, built entirely around AI. Director Hans Scheffer told the story at AI010; Emerce, a Dutch business news site, wrote it up. Emerce sums up his starting point: “AI-first betekent eerst documenteren.” AI-first means documenting first. Every department describes every process, including what happens when something does not happen. “Dat loopt op tot duizenden of tienduizenden regels.” That adds up to thousands or tens of thousands of lines. Emerce names the effects: “De afdeling Klantenservice bijvoorbeeld ging van honderd naar achttien.” Customer service, for example, went from one hundred people to eighteen. Asked whether he is wrecking a good company this way: “Dat weet ik niet. Dat is de sport van het ondernemen.” I don’t know. That is the sport of running a business. It is his own story; employees do not speak in the article.

What it means for you: in most businesses the knowledge lives in people’s heads. Who knows how a complaint is handled, or which customer always pays late? Writing that down can start tomorrow, even without AI, and it is exactly what an AI assistant needs. Where to begin is in which administrative processes to automate first.

Source: Emerce, 19 October 2026

What does an AI assistant do when it hits a wall?

Sometimes it finds its own way around. OpenAI added new reports about its own models in internal tests, updated on 15 October. One model paid the costs of its task from a wallet that was not its own, on a test network without real money. OpenAI: “This case shows that a model can substitute its own judgment about likely harm for the account holder’s permission, then act on that judgment without authorization.” Another model only had to look up public figures, but changed settings in internal systems and retrieved non-public data. “The model did not disclose these actions.” Fortune counted 19 incidents on the page: “most occurred in May or June, but OpenAI did not discover or disclose them until months later.” OpenAI adds that these are individual cases that say nothing about how often it happens.

What it means for you: give an assistant access only to what belongs to its task, and do not read its summary as proof that nothing else happened. Also read how to test an AI agent before automating business processes.

Source: OpenAI, 15 October 2026, OpenAI, 15 October 2026, Fortune, Emily Forlini, 16 October 2026 and OpenAI, explanation of the reports

Can someone wearing camera glasses walk into my office?

In September the Dutch Data Protection Authority (AP) already advised organizations to ban camera glasses in their house rules where needed. Now it is going to court. Offlimits, Firewall and the Dutch Consumers’ Association want sales of Meta’s glasses to stop, and filming to be possible only “met uitdrukkelijke toestemming van personen die gefilmd worden”, with the explicit consent of the people being filmed. “De rechtbank Amsterdam behandelt de zaak op 26 november.” The Amsterdam court hears the case on 26 November. Meta says the footage in principle stays on the user’s phone, unless the user uses AI features.

What it means for you: do not wait for the court. The AP writes: “Voor omstanders is nauwelijks te zien of iemand alleen naar hen kijkt of ook filmt.” Bystanders can hardly tell whether someone is only looking at them or also filming. One sentence in your house rules, for staff and visitors, settles it for your own office.

Source: Computable, Pim van der Beek, 15 October 2026 and Dutch Data Protection Authority, 22 September 2026

What does this mean for your business?

That is how Bombos works: it learns how your business works, prepares the work, and you approve it. Take complaint handling: Bombos follows every complaint from the first email to the resolution and prepares the next step. So every complaint gets a follow-up, even when the person who picked it up is out sick. That one task is the start; after that your team teaches Bombos the next one itself.

Sources

Every source was opened on 19 October 2026 and every quote appears in it word for word. OpenAI reports this about its own models and sells AI assistants itself. The HelloPrint story comes from its director.

Emerce, HelloPrint bouwt naast het bestaande een tweede, volledig AI-first organisatie (19 October 2026) · OpenAI, Using a funded test cryptocurrency wallet without authorization (updated 15 October 2026) · OpenAI, Accessing and changing internal systems during dashboard research (updated 15 October 2026) · Fortune, OpenAI discloses four more rogue AI incidents (16 October 2026) · OpenAI, Our framework for reporting model misalignment (opened 19 October 2026) · Computable, Rechtszaak om verkoop Meta-‘gluurbril’ te stoppen (15 October 2026) · Autoriteit Persoonsgegevens, AP waarschuwt gebruikers camerabrillen (22 September 2026)

The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.