With Data Privacy Day coming up, we spoke to cybersecurity industry experts about the latest data privacy trends.
Many early AI services and tools, including ChatGPT, employ a usage model that's similar to social media services like Facebook and TikTok.
While you don't pay money to use those platforms, you are compensating them through the sharing of your private data, which these companies leverage and monetise through ad targeting.
A free AI service can collect data from your devices and store your prompts, then use that data to train its own model.
While this may not seem malicious, it's precisely why it's so crucial to analyse the privacy implications of processing scraped data to train generative AI algorithms.
Say one of these companies gets breached; threat actors could obtain access to your data, and - just like that - have the power to weaponize it against you.
AI models have huge vulnerabilities, and those vulnerabilities are very difficult to fix.
Read the fine print of your AI provider's policies.
OpenAI claims that they will not use enterprise customers' data to train their models.
That doesn't protect you from hostile attacks that might leak your data, but it's a big step forward.
Other providers will eventually be forced to offer similar protections.
Don't say anything to an AI that you wouldn't want leaked.
It's possible to run several open source models entirely on your laptop; no cloud, no Internet required once you've downloaded the software.
The performance of these models isn't quite the equal of the latest GPT, but it's impressive.
Llamafile is the easiest way to run a model locally.
The transformational factor of 2023 was, of course, the explosion of Generative AI onto the world.
Over the course of the year, we were bombarded with more and more innovative and amazing examples of content generated by these systems but were also increasingly educated in the often cavaliere way these systems were greedily 'trained' on every piece of human generated content imaginable.
From our family photos on social media, to fan-fiction stories from niche internet forums, a range of organisations grabbed all the 'publicly available' data they could from the internet with abandon.
Over the past decade, society has become more and more comfortable in trading data about our private lives, preferences, thoughts, and experiences, to get everything from tailored shopping and streaming media recommendations to cheaper insurance premiums or customised healthcare coverage.
LLMs and other Generative AI systems are now not only optimizing the content and products we consume, but constructing or hallucinating them from whole cloth, with often blurry lines in terms of creative ownership of that content.
This Cyber News was published on www.itsecurityguru.org. Publication date: Fri, 26 Jan 2024 17:13:04 +0000