// privacy

Source: The Next Web

What started as an academic curiosity—hiding instructions in invisible Unicode characters—has entered the operational toolkit of commercial spam campaigns. Security researchers publish findings on AI model weaknesses, then low-skill actors immediately commodify them for scale, forcing filter maintainers into perpetual catch-up. For platforms and email providers, this is less a technical problem than a resource one. Fighting invisible character obfuscation requires constant ruleset updates, while spammers can iterate trivially by adding new Unicode combinations.

Source: 404 Media

Amazon is running an industrial-scale operation that pulps purchased books to extract training data for its AI systems—a direct pipeline from physical inventory to machine learning infrastructure that sidesteps licensing negotiations entirely. Tech companies are now treating intellectual property as raw material: rather than negotiating with publishers or paying for content rights, Amazon converts physical books into training data. The practice exposes the economic logic of the AI boom: training models at scale requires such vast quantities of text that legal constraints on data sourcing become obstacles to be engineered around, not rules to follow.

Source: Beet

With fewer than 30% of consumers identifiable through traditional tracking methods, the ad tech infrastructure built over two decades is functionally obsolete. Publishers and platforms are abandoning deterministic targeting for contextual, cohort-based, and first-party strategies. The addressability that made programmatic advertising viable is collapsing now, not gradually. Companies betting on cookie alternatives and identity solutions face a market where most consumer interactions remain opaque. The winners will monetize anonymity, not track better.