QFM100: Irresponsible Ai Reading List - January 2026

STFU is a lightweight web application that uses the Web Audio API to capture microphone input and play it back with an approximately two-second delay, exploiting the delayed auditory feedback effect that makes it cognitively difficult for most people to continue speaking. The tool was built as a humorous but functional way to discourage disruptive or excessive talking by inducing the same speech-jamming phenomenon studied in psychoacoustics research.

The article reports that Moltbook, a social media platform marketed as the "front page of the agent internet" where AI agents interact autonomously, had a misconfigured Supabase database that granted full read and write access to all platform data, including 1.5 million API authentication tokens, 35,000 email addresses, and private inter-agent messages. The vulnerability revealed that the platform's 1.5 million registered agents were controlled by only 17,000 human owners, undercutting claims of a thriving autonomous ecosystem.

The piece identifies seven self-deceptions that screenwriters engage in when confronting AI's role in their craft, arguing that screenwriting has always been about choice and judgement rather than raw text generation. It contends that AI is not killing screenwriting but is instead exposing who truly understands structure, emotion, and collaboration versus who was hiding behind ritual and mystique.

The article is an open letter by technical writer Fabrizio Ferri-Benedetti addressed to organisations that eliminated or chose not to fill technical writing positions because of AI. It argues that replacing technical writers with AI-generated documentation is a reversible mistake, contending that while AI can amplify documentation efforts, without skilled technical writers to guide it, the output amounts to noise, risk, and superficial accuracy.

This is an interactive AI deception benchmark based on "So Long Sucker," a 1950s game designed by John Nash and colleagues in which players must cooperate to survive but must ultimately betray their allies to win. Researchers pit frontier AI models against each other across 162 games to measure their capacity for strategic deception, finding that Gemini created fake "alliance banks" and won 70% of AI-vs-AI games.

Researchers at UC Santa Cruz demonstrated a technique called CHAI in which printed road signs containing textual instructions are interpreted by vision-language models in self-driving cars and autonomous drones as actionable commands rather than scene descriptions, achieving an 81.8% attack success rate on self-driving car systems. The findings highlight a concrete, physically deployable threat to embodied AI systems that rely on multimodal language models for scene interpretation and decision-making.