Szkolenie z etycznej sztucznej inteligencji: Jak Oxylabs na nowo definiuje zgodę w erze sztucznej inteligencji

Step inside Lithuania’s vibrant tech landscape, and you’ll likely come across Denas Grybauskas of Oxylabs. As the company’s Chief Governance and Strategy Officer, he doesn’t shy away from the tough stuff—instead, he leads conversations around the ethical boundaries of artificial intelligence at a time when few others want to wrestle with the topic.

What drives Grybauskas is a commitment to responsible progress. He’s adamant that innovation in AI shouldn’t leave accountability in the dust. “Ethical practices are deeply rooted in how we operate,” Grybauskas shares. “Whether it’s making sure our proxy networks set a new ethical standard or ensuring that AI training data is responsibly sourced—our approach always comes back to integrity.”

Since its founding in 2015, Oxylabs has carved out a reputation for pioneering projects. The company boasts one of the world’s largest ethically sourced proxy networks, reaching 195 countries and totaling more than 177 million unique IPs. Their latest achievement takes things a step further: Oxylabs introduced what they say is the world’s first YouTube datasets for AI training built entirely on creator consent. Every dataset, from the videos themselves to associated transcripts and metadata, is shared only with the clear approval of each content creator. In Grybauskas’ words: “We’re not just collecting data. We’re building bridges between the creators of content and those fueling innovation, always with mutual respect at the heart.”

He’s convinced this idea of consent-driven data can go much further. Grybauskas envisions a future where everything from music to literature to digital art is systematically, simply, and fairly licensed for use in AI—creating an ecosystem where everyone benefits and scaling up doesn’t mean compromising on principle.

But bringing this to life is a constant balancing act. Negotiating between what’s fair for creators and how quickly AI can develop is no easy road. Grybauskas points to ongoing debates, especially in the UK, highlighting why it’s crucial to protect creator rights. “Yes, innovation is essential,” he says, “but not at the creator’s expense. They deserve to feel secure, not taken advantage of. Regulation must keep pace and protect that balance.”

One criticism he often hears is that seeking consent might slow down progress. Grybauskas strongly disagrees. “We don’t see an either/or situation between innovation and ethics. We’ve proven it’s possible to have smart, scalable infrastructure that makes consent straightforward, not a hurdle. Our datasets are an example that robust tech can support both sides.”

Of course, pushing boundaries presents its own technical headaches. Structuring large amounts of highly varied data into workable pieces wasn’t simple for Oxylabs, Grybauskas admits. But with years of experience and a solid backbone of technical know-how, the team managed to pull it off.

For Grybauskas, the ethical approach isn’t just about doing the right thing—it’s also a way to bring more consistency to patchwork AI regulations across different countries. Simplifying how consent is obtained and honored could set a global standard, making ethical best practices the norm rather than the outlier.

He’s equally passionate about the need for openness. Grybauskas believes that being transparent about how AI is trained isn’t a vulnerability—it’s a foundation for rebuilding trust, which in turn paves the way for AI’s long-term acceptance and success. It’s a belief reflected in initiatives like Oxylabs’ Ethical Web Data Collection Initiative and Project 4β, all focused on setting—not just following—the new rules of the game.

His final piece of advice for those entering the AI space? Don’t see ethical data as a roadblock. In fact, Grybauskas is adamant that it’s quickly becoming a competitive edge for those ready to embrace it.

For deeper insight into how Oxylabs is putting all this into practice, including the full interview with Denas Grybauskas, you can find the original article at Unite.AI.