Experts back watermarks to curb excessive reliance on AI

Dar es Salaam. The introduction of digital watermarks to identify content generated by artificial intelligence (AI) has received backing from Tanzanian technology, academic and creative industry experts, who say the move could restore accountability and encourage people to develop their own skills rather than rely excessively on machines.

Major AI companies such as Anthropic, the maker of Claude, OpenAI, the maker of ChatGPT and Google are developing ways to identify AI-generated content. Anthropic has introduced watermarking for text generated by its new Claude models, while Google uses SynthID to watermark AI-generated images.

Anthropic began marking text generated by new Claude models launched from August 2, 2026. The company says the watermark is imperceptible, travels with text when copied and pasted and can survive some editing.

Similar technologies are being explored for video and audio amid growing concerns over misinformation, copyright and the presentation of AI-generated work as entirely human-created.

The move also follows regulatory pressure in Europe, where transparency requirements under the EU AI Act took effect on August 2, 2026, requiring AI providers to make certain AI-generated content identifiable.

However, watermarking is not foolproof. Anthropic says a watermark can indicate that Claude was involved with content but cannot establish that the AI was its original author. Heavy editing, paraphrasing or translation can also affect detection.

In Tanzania, Plate AI co-founder Janeth-Kareen Kilonzo said the development was important because it would ensure people received credit for work they had actually produced.

“It is good because everyone will get the credit they deserve. There are times when we credit people beyond the reality of their ability,” she said.

Ms Kilonzo said disclosure was particularly important as some people increasingly use chatbots to produce substantial portions of their work while presenting the final product as their own.

“We are involved in meal planning through an AI system, but it clearly says, even in its name, that it is AI,” she said.

The problem, she said, was that some individuals use AI extensively without disclosing its involvement.

“Some people are having a lot done for them by AI chatbots but they don't say so. Instead, they present it as work they have done through their own effort. Intellectually, that is theft. It is only because they are chatbots that we don't see people being taken to court,” she said.

Chief Executive Officer of Tanzania-based AI technology company Sartify, Dr Michael Mollel, said the development could help society establish a healthier relationship with AI.

He said people should regard AI as an assistant rather than allow it to replace their thinking, creativity and skills.

“This will help people put more effort into producing better things and use AI as an assistant, rather than over-relying on AI,” Dr Mollel said.

For the creative industry, music producer Salmin Maengo, popularly known as S2Kizzy, said AI-generated work was becoming increasingly common, although experts could often identify weaknesses in machine-generated content.

“In music, we look at how a song sounds. Sometimes you listen and realise there is no point where the person has breathed from the beginning to the end of the song,” he said.

Mr Maengo said watermarking could benefit the creative industry by encouraging people who have concentrated mainly on prompting AI systems to acquire genuine creative skills.

“This will help the creative industry continue to innovate. Even those who wanted to remain in the business of simply prompting things will be pushed to learn real skills because they will know AI itself can expose that it produced the work,” he said.

At Ardhi University, lecturer Ms Jasmine Christian said AI identification could also help address challenges in education, where lecturers increasingly struggle to establish whether assignments reflect students' own work.

“This will help greatly in academia because it has become difficult to know whether this is genuinely a person's work or AI,” she said.

However, Ms Christian acknowledged that watermarking alone would not eliminate dishonest use of AI.

“Even with this, there are still options where people can prompt and then reproduce what was typed and it can still appear the same,” she said.

Despite the limitation, she said the development could help shift society towards valuing human thinking and intellectual effort.

“At least it shows an intention to return the world to people thinking, rather than leaving machines to do the work for them,” she said.