Elon Musk also recently took to his X account to announce that xAI’s chatbot is coming to Teslas. The announcement came after quite a rough week for Grok, which experienced a sort of meltdown, praising Adolf Hitler and instructing users on how to commit sexual assault.
In the early years of large language models (LLMs), when discussing language models vs. artificial general intelligence (AGI), I remember people joking that you wouldn’t want your chatbot driving your Tesla. It’s not funny anymore. The sight of people saying, “Grok, park my car and keep it cool till I come back,” is probably not that far away.
Karolis Arbaciauskas, Head of Business Product at NordPass further discusses.
Agents and passwords
It’s only a matter of time before our aspirations to further empower AI agents emerge. The use case where AI agents use password managers and even banking apps on behalf of the user is probably in the very near future. Prompt “calculate and pay the utility bills while I go for a run” sounds appealing, doesn’t it?
In principle, we can already send agents to password vaults, allow them to retrieve passwords, and perform certain operations. There are ways to do that, and they work. However, at this point it is extremely unsafe.
In the near future, AI agents (operators) will likely be able to retrieve passwords or other secrets from password vaults through API integrations without compromising their own login credentials. Such a model of machine-to-machine authentication is already working in other scenarios. It is also secure in principle. The only questions are how much control will the AI have and if it or threat actors will be able to somehow exploit this access further?
We were promised robots but got social networks instead
Do we want this to happen? I think we do. Pop culture – especially books, movies, and games – has long created expectations for this. And in recent years businesses, with the help of the media, have been fuelling these expectations. So, people in general, or should I say we as a humanity, seem to be waiting for AGI, even though we worry about our privacy and are a little afraid of it. Agentic AI is the closest thing we have right now, so I’m sure the technology will catch on and evolve further.
Especially seeing how much money venture capital is pouring into AI startups. According to PitchBook, in the first half of 2025 more than half of all venture capital dollars globally, and 64% in the US, went to AI startups. Over the same period, AI helped 36 tech companies achieve unicorn status.
I won’t go into technology adoption theories (such as Diffusion of innovation or TAM), but KPMG is right in saying that agentic AI deployment will accelerate despite its risks. Why? Because if businesses want it, and people want it, it will happen. We just need to be careful about potential vulnerabilities and how much control we give away to AI agents. We still don’t know what might happen when the real AGI emerges.
Let’s not forget that passwords to all our accounts (via access to password managers) and banking data are among the most important and most valuable, to us, to AI agents (because when we give them access to our credentials, their capabilities grow significantly), and to criminals. At the same time, the metadata of our interactions with AI agents is very valuable for companies that created those agents.
