Jan 09, 2026 20:00:00

Firefox fork 'Waterfox' developer criticizes Mozilla's AI policy

In December 2025, Anthony Enzo DeMeo, who led the Firefox development team, became the new CEO of Mozilla and announced a policy of becoming the world's most trusted software company with AI at its core. Alex Kontos, creator of the Firefox fork '

Waterfox ,' objected to this policy, saying that 'the company is pursuing fundamentally untrustworthy technology.'

No AI* Here - A Response to Mozilla's Next Chapter - Waterfox Blog
https://www.waterfox.com/blog/no-ai-here-response-to-mozilla/

Enzo DeMeo, who has been appointed CEO, said, 'Our top priority remains building the best browser, which is our core business,' and outlined his intention to incorporate AI features into the browser experience.

Mozilla's new CEO talks about the future of Firefox, prioritizing the 'best browser' while also integrating AI - GIGAZINE

Kontos said that while he understands Firefox's move to incorporate AI, given that browsers with AI are beginning to take over the market, he is opposed to introducing a black box system whose behavior cannot be objectively verified. He said that simple AI functions are fine, but he is skeptical of the introduction of agent-type AI that performs a wide range of functions.

For example, a translation function can maintain a certain degree of objectivity because it simply translates text. However, agent AI that acts on behalf of the user and performs all sorts of actions is fundamentally untrustworthy, as it 'determines what the user sees and how they see it based on logic that cannot be verified or understood,' Kontos argues.

In fact, the behavior of large-scale language models is opaque and black-boxed, making it impossible to objectively verify their behavior from the outside. Even if we were to give a prompt to 'describe the assumptions about the behavior,' it is known that the assumptions output do not accurately reflect the actual behavior.

Anthropic tests its Claude 3.7 Sonnet and DeepSeek-R1 inference models that output 'thought content' but are inconsistent with actual thought content - GIGAZINE

Kontos pointed out, 'How do you know what a black-box AI is doing, or even if it's on? How do you audit it? How can you be sure that it's not reshaping your browsing experience without you even realizing it?'

'Mozilla touts trust, transparency, and user autonomy, while at the same time actively embracing technologies that undermine these three principles. The core browsing experience should be one that users have complete control over, not one that claims to help them but constantly monitors them through mysterious systems,' he continued.

Waterfox is designed for users who want it to function simply as a browser, and will not include features that utilize large-scale language models in the future.