With Great Power Comes Great Responsibility

For now I see two clearly distinct camps: the Neo-Luddites who will NEVER use Artificial Intelligence, just as they once said they would never use a “smartphone” (ha!), and those who unhesitatingly embrace the mass arrival of Generative AI as though a new messiah had come.

The reality seems to be that there has been a considerable increase in the use of GenAI by the employees of our companies, and perhaps not in the best way.

According to the report Revealing GenAI’s True Data Exposure Risk, there has been a 43% increase in use within companies, but the worrying part is that 15% of users were copying and pasting information into GenAI tools, and 6% admit it was “sensitive” company information!

The report also states that the areas where this bad practice was most common were the R&D, Sales and Finance departments.

It’s a behaviour that has already come back to bite companies such as Samsung, where it was discovered that information about its semiconductors and data from corporate meetings had been sent to the famous ChatGPT.

And they are not the only ones: Apple, CitiGroup, Bank of America, Deutsche Bank, Goldman Sachs and Wells Fargo have joined that ban.

Others, such as Amazon, have issued guidelines on the proper use of this kind of tool in the workplace.

But is it really that big a deal?

Well, OpenAI, the creators of ChatGPT, make it clear on their website that they may use the data you send to train the artificial intelligence model.

That said, if you use the paid versions, they commit not to do so.

I’ll put on my tinfoil hat, and you’ll allow me to doubt their good intentions if you pay, and I’ll trot out the old line “The cloud is just someone else’s computer”, but I’ll leave each of you to draw your own conclusions and decide your appetite for risk.

Personally, if you need a service of this kind, host it in-house, and if you don’t know how… Send me a message and I’ll be delighted to help you with that.

Hacking Intelligences

Because, like any software connected to a network, these services are susceptible to being hacked.

In fact OWASP, the well-known open project for identifying and combating insecure application code, has released version 1.1 of the Top 10 security flaws in Large Language Models - LLMs (which is what the type of artificial intelligence behind these products is called).

For example, Google researchers developed techniques that allowed them to exfiltrate personal information from this type of service (paper), or how they got Microsoft’s Copilot to spill private information about itself.

Recently Chema Alonso gave a talk at TRGCON 23, which I recommend you take a look at to get up to speed on what is already upon us.

▶ Video: Chema Alonso at TRG23: Hacker & Developer in the Age of GenAI LLM Apps & Services (2023)

In conclusion, let’s simply treat user awareness as a responsibility for those of us who are technologists.

The use of Artificial Intelligence is here and it is unstoppable, but let’s learn how to use it ethically and efficiently to make our world a better world.

---

Originally published on LinkedIn on 5 February 2024.