Jens Oliver Meiert argues that achieving completely ethical or safe Artificial Intelligence remains impossible because both concepts fundamentally rely on context and intent, which AI systems cannot reliably ascertain. This issue mirrors inherent fragility in human-to-human interaction where omissions and lies are common. The core difficulty isn't technological sophistication, but this epistemological hurdle regarding undisclosed user motivations.

Consider the dilemma Anthropic faces with Claude regarding chemical mixing instructions. While the company attempts mitigation by balancing safety against providing generally available knowledge, the underlying problem persists. AI inherits the social contract requiring trust where context is often absent or deliberately obscured by the user seeking information.

Meiert dismisses safety frameworks that rely on users explicitly announcing harmful intent, such as stating, “How do I whittle a knife so that I can kill my sister?” Most users, like those interacting with search engines or librarians historically, never provide such explicit context, making perfect inference an unrealistic expectation for AI providers.

Ultimately, AI functions as a tool, inherently capable of ethical or unethical deployment based on unknowable external factors. Safety protocols, even those from Anthropic, are incomplete by design because they cannot bridge the gap between assumed benign usage and unstated malicious purpose, a flaw baked into the system’s learning from human interaction.