AI pioneers warn of ‘vibe slop’ flooding software with faulty code

09:05

By: Dakir Madiha

AI pioneers warn of ‘vibe slop’ flooding software with faulty code

Engineers behind the open source AI agent system OpenClaw are warning about what they describe as “vibe slop,” a growing wave of low-quality and potentially unsafe code generated by artificial intelligence tools. The concern comes as AI-assisted programming rapidly becomes standard practice across the technology industry, transforming how software is written and maintained.

The expression combines “vibe coding,” a term coined last year by Andrej Karpathy to describe generating software through natural language prompts, with “AI slop,” a label used for poor-quality machine-generated content spreading online. Developers involved in the debate say the problem emerges when programmers replace structured engineering practices such as testing, architecture review and security validation with direct AI-generated outputs that receive little human oversight.

OpenClaw, previously known as MoltBot and Clawdbot, has become a central example in the security debate surrounding autonomous coding agents. The framework gives AI agents system-level permissions to run commands, edit files and connect with outside services. Security researchers identified more than 140,000 publicly exposed OpenClaw instances online, many of which reportedly leaked API keys and login credentials. Several high-severity vulnerabilities have been disclosed since January, prompting companies including Meta to limit its internal use.

OpenClaw creator Peter Steinberger had previously criticized the phrase “vibe coding,” calling it an insult during an interview earlier this year. Engineers now raising concerns about “vibe slop” argue that the deeper issue lies in inexperienced users relying on AI systems to produce software they cannot properly audit, debug or maintain. Critics warn this could introduce large volumes of insecure and poorly documented code into production systems.

The debate is spreading beyond software development into scientific research and academic publishing. On May 21, the chief technology officer of an AI-assisted research startup warned about the rise of “hypothesis slop,” describing a growing influx of AI-generated scientific work containing fabricated or unverifiable information. An April analysis found submissions to a major management journal increased by 42 percent after the launch of OpenAI’s ChatGPT, with AI-heavy manuscripts proving harder to read and more likely to face rejection. Researchers and conference organizers are also struggling with a rise in fully AI-generated papers containing invented citations and false data.

The warning arrives as companies already face mounting technical debt inside aging software systems, a burden estimated at more than $1.5 trillion globally. Industry experts say uncontrolled adoption of AI-generated code could accelerate that problem by creating vast new layers of software that organizations neither fully understand nor properly secure.